tempest-react-sdk 0.36.0 → 0.38.0

package/dist/vision/tasks/classifier.js

@@ -1,21 +1,23 @@
 import { SpeedTimer as e } from "../core/timing.js";
 import { ClassificationResults as t, Probs as n } from "../results.js";
 import { resolveLabels as r } from "../labels.js";
-import { OrtSession as i } from "../core/session.js";
-import { loadImage as a } from "../io/image.js";
-import { normalize as o, resize as s, toCHW as c, toFloat32Tensor as l } from "../preprocess/image.js";
-import { softmax as u, topK as d } from "../postprocess/classification.js";
-import { VisionTask as f } from "./base.js";
+import { classificationNumClasses as i, resolveInputSize as a } from "../core/graph.js";
+import { modelNames as o } from "../core/metadata.js";
+import { OrtSession as s } from "../core/session.js";
+import { loadImage as c } from "../io/image.js";
+import { normalize as l, resize as u, toCHW as d, toFloat32Tensor as f } from "../preprocess/image.js";
+import { softmax as p, topK as m } from "../postprocess/classification.js";
+import { VisionTask as h } from "./base.js";
 //#region src/vision/tasks/classifier.ts
-var p = [
+var g = [
 	.485,
 	.456,
 	.406
-], m = [
+], _ = [
 	.229,
 	.224,
 	.225
-], h = class h extends f {
+], v = class v extends h {
 	_labels;
 	_names;
 	_inputSize;
@@ -25,10 +27,14 @@ var p = [
 	constructor(e, t, n, r, i, a, o) {
 		super(e), this._labels = t, this._names = n, this._inputSize = r, this._mean = i, this._std = a, this._applySoftmax = o;
 	}
-	static async create(e, t) {
-		let n = await i.create(e, t), a = r(t.labels, { numClasses: t.numClasses }), o = {};
-		for (let e = 0; e < a.length; e++) o[e] = a[e];
-		return new h(n, a, o, t.inputSize ?? [224, 224], t.mean ?? p, t.std ?? m, t.applySoftmax ?? !0);
+	static async create(e, t = {}) {
+		let n = await s.create(e, t), c = t.numClasses ?? i(n.outputShape) ?? void 0, l = r(t.labels ?? o(n.metadata), { numClasses: c }), u = {};
+		for (let e = 0; e < l.length; e++) u[e] = l[e];
+		return new v(n, l, u, a({
+			graphShape: n.inputShape,
+			requested: t.inputSize,
+			fallback: [224, 224]
+		}), t.mean ?? g, t.std ?? _, t.applySoftmax ?? !0);
 	}
 	get labels() {
 		return this._labels;
@@ -36,6 +42,9 @@ var p = [
 	get names() {
 		return this._names;
 	}
+	get inputSize() {
+		return this._inputSize;
+	}
 	get numClasses() {
 		return this._labels.length;
 	}
@@ -43,17 +52,17 @@ var p = [
 		return this.predict(e, t);
 	}
 	async predict(r, i = {}) {
-		let o = new e(), s = typeof r == "string" ? r : null, c = await a(r);
-		o.stage("load");
-		let l = this._preprocess(c);
-		o.stage("preprocess");
+		let a = new e(), o = typeof r == "string" ? r : null, s = await c(r);
+		a.stage("load");
+		let l = this._preprocess(s);
+		a.stage("preprocess");
 		let u = await this._session.run({ [this._session.inputName]: l });
-		o.stage("inference");
-		let f = this._session.outputNames[0];
-		if (f === void 0) throw Error("Classifier model has no outputs.");
-		let p = u[f];
-		if (p === void 0) throw Error(`Classifier model output ${f} missing from run() result.`);
-		let m = this._postprocess(p.data), { indices: h, values: g } = d(m, i.topK ?? null), _ = [];
+		a.stage("inference");
+		let d = this._session.outputNames[0];
+		if (d === void 0) throw Error("Classifier model has no outputs.");
+		let f = u[d];
+		if (f === void 0) throw Error(`Classifier model output ${d} missing from run() result.`);
+		let p = this._postprocess(f.data), { indices: h, values: g } = m(p, i.topK ?? null), _ = [];
 		for (let e = 0; e < h.length; e++) {
 			let t = h[e], n = this._labels[t] ?? `class_${t}`;
 			_.push({
@@ -73,14 +82,14 @@ var p = [
 			cls: v.classId,
 			name: v.className,
 			conf: v.probability,
-			image: c,
+			image: s,
 			probabilities: _
-		}, b = [c.height, c.width], x = new n(m);
-		return o.stage("postprocess"), [new t(x, y, this._names, c, b, s, o.speed())];
+		}, b = [s.height, s.width], x = new n(p);
+		return a.stage("postprocess"), [new t(x, y, this._names, s, b, o, a.speed())];
 	}
 	_preprocess(e) {
-		let [t, n] = this._inputSize, r = s(e, t, n);
-		return l(c(o(r, this._mean, this._std), r.width, r.height, 3), [
+		let [t, n] = this._inputSize, r = u(e, t, n);
+		return f(d(l(r, this._mean, this._std), r.width, r.height, 3), [
 			1,
 			3,
 			r.height,
@@ -88,10 +97,10 @@ var p = [
 		]);
 	}
 	_postprocess(e) {
-		return this._applySoftmax ? u(e) : new Float32Array(e);
+		return this._applySoftmax ? p(e) : new Float32Array(e);
 	}
 };
 //#endregion
-export { h as Classifier };
+export { v as Classifier };
 
 //# sourceMappingURL=classifier.js.map

package/dist/vision/tasks/classifier.js.map

@@ -1 +1 @@
-{"version":3,"file":"classifier.js","names":[],"sources":["../../../src/vision/tasks/classifier.ts"],"sourcesContent":["/**\n * Image classification task using ONNX Runtime Web.\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { softmax, topK } from \"../postprocess/classification\";\nimport { normalize, resize, toCHW, toFloat32Tensor } from \"../preprocess/image\";\nimport { ClassificationResults, Probs } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type ClassProbability, type ClassificationResult, type RGBImage } from \"../types\";\n\nconst IMAGENET_MEAN: readonly [number, number, number] = [0.485, 0.456, 0.406];\nconst IMAGENET_STD: readonly [number, number, number] = [0.229, 0.224, 0.225];\n\nexport interface ClassifierOptions extends OrtSessionOptions {\n    /** Class label spec — see {@link resolveLabels}. */\n    readonly labels: LabelSpec;\n    /**\n     * Number of classes the model can predict. Required when `labels` is `null`\n     * or when you want to validate that the supplied labels match the model.\n     */\n    readonly numClasses?: number;\n    /** Model input `[width, height]` in pixels. Defaults to `[224, 224]`. */\n    readonly inputSize?: readonly [number, number];\n    /** Per-channel RGB mean used for normalization. Defaults to ImageNet. */\n    readonly mean?: readonly [number, number, number];\n    /** Per-channel RGB standard deviation. Defaults to ImageNet. */\n    readonly std?: readonly [number, number, number];\n    /**\n     * If `true` (default), apply softmax to the raw model output. Set to\n     * `false` for models whose final layer already produces a probability\n     * distribution.\n     */\n    readonly applySoftmax?: boolean;\n}\n\nexport interface ClassifierPredictOptions {\n    /**\n     * If set, the per-class probability list in `results[0].result.probabilities`\n     * is truncated to the top-K entries. The bulk `probs` view always exposes\n     * the full vector.\n     */\n    readonly topK?: number;\n}\n\n/**\n * Image classifier wrapping an ONNX model with ImageNet-style preprocessing.\n *\n * `predict()` returns `Promise<ClassificationResults[]>` (length 1 for a\n * single image), mirroring Ultralytics' API. The envelope exposes a `probs`\n * collection (`top1`, `top1conf`, `top5`, `top5conf`, `data`) plus the\n * legacy per-class probability list with names already resolved.\n *\n * Defaults: 224×224 RGB input, `float32` normalized with ImageNet mean/std,\n * NCHW layout, batch size 1, softmax applied to the raw output.\n *\n * @example\n * ```typescript\n * const clf = await Classifier.create(\"/models/resnet50.onnx\", {\n *   labels: [\"tench\", \"goldfish\", ...]  // 1000 ImageNet labels\n * });\n * const r = (await clf.predict(\"/images/dog.jpg\"))[0];\n * console.log(r.cls, r.conf, r.name);\n * console.log(r.probs.top5, r.probs.top5conf);\n * ```\n */\nexport class Classifier extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _mean: readonly [number, number, number],\n        private readonly _std: readonly [number, number, number],\n        private readonly _applySoftmax: boolean,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: ClassifierOptions): Promise<Classifier> {\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels, { numClasses: options.numClasses });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Classifier(\n            session,\n            labels,\n            names,\n            options.inputSize ?? [224, 224],\n            options.mean ?? IMAGENET_MEAN,\n            options.std ?? IMAGENET_STD,\n            options.applySoftmax ?? true,\n        );\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /** Number of classes the model can predict. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /** Alias for {@link predict} (parity with PyTorch `nn.Module.__call__`). */\n    async call(\n        image: ImageInput,\n        options: ClassifierPredictOptions = {},\n    ): Promise<ClassificationResults[]> {\n        return this.predict(image, options);\n    }\n\n    /** Run classification on a single image. */\n    async predict(\n        image: ImageInput,\n        options: ClassifierPredictOptions = {},\n    ): Promise<ClassificationResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const tensor = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Classifier model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(\n                `Classifier model output ${firstOutputName} missing from run() result.`,\n            );\n        }\n        const fullProbs = this._postprocess(raw.data as Float32Array);\n\n        const { indices, values } = topK(fullProbs, options.topK ?? null);\n        const probabilities: ClassProbability[] = [];\n        for (let i = 0; i < indices.length; i++) {\n            const id = indices[i] as number;\n            const className = this._labels[id] ?? `class_${id}`;\n            probabilities.push({\n                classId: id,\n                className,\n                probability: values[i] as number,\n                cls: id,\n                name: className,\n                conf: values[i] as number,\n            });\n        }\n        if (probabilities.length === 0) {\n            throw new Error(\"Classifier produced no probabilities (empty output).\");\n        }\n\n        const top = probabilities[0] as ClassProbability;\n        const result: ClassificationResult = {\n            classId: top.classId,\n            className: top.className,\n            confidence: top.probability,\n            cls: top.classId,\n            name: top.className,\n            conf: top.probability,\n            image: original,\n            probabilities,\n        };\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const probs = new Probs(fullProbs);\n        timer.stage(\"postprocess\");\n        return [\n            new ClassificationResults(\n                probs,\n                result,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): ort.Tensor {\n        const [tw, th] = this._inputSize;\n        const resized = resize(image, tw, th);\n        const normalized = normalize(resized, this._mean, this._std);\n        const chw = toCHW(normalized, resized.width, resized.height, 3);\n        return toFloat32Tensor(chw, [1, 3, resized.height, resized.width]);\n    }\n\n    private _postprocess(raw: Float32Array): Float32Array {\n        return this._applySoftmax ? softmax(raw) : new Float32Array(raw);\n    }\n}\n"],"mappings":";;;;;;;;;AAgBA,IAAM,IAAmD;CAAC;CAAO;CAAO;AAAK,GACvE,IAAkD;CAAC;CAAO;CAAO;AAAK,GAsD/D,IAAb,MAAa,UAAmB,EAAW;CAGlB;CACA;CACA;CACA;CACA;CACA;CAPrB,YACI,GACA,GACA,GACA,GACA,GACA,GACA,GACF;EADmB,AAEjB,MAAM,CAAO,GAPI,KAAA,UAAA,GACA,KAAA,SAAA,GACA,KAAA,aAAA,GACA,KAAA,QAAA,GACA,KAAA,OAAA,GACA,KAAA,gBAAA;CAGrB;CAGA,aAAa,OAAO,GAAoB,GAAiD;EACrF,IAAM,IAAU,MAAM,EAAW,OAAO,GAAO,CAAO,GAChD,IAAS,EAAc,EAAQ,QAAQ,EAAE,YAAY,EAAQ,WAAW,CAAC,GACzE,IAAgC,CAAC;EACvC,KAAK,IAAI,IAAI,GAAG,IAAI,EAAO,QAAQ,KAC/B,EAAM,KAAK,EAAO;EAEtB,OAAO,IAAI,EACP,GACA,GACA,GACA,EAAQ,aAAa,CAAC,KAAK,GAAG,GAC9B,EAAQ,QAAQ,GAChB,EAAQ,OAAO,GACf,EAAQ,gBAAgB,EAC5B;CACJ;CAGA,IAAI,SAA4B;EAC5B,OAAO,KAAK;CAChB;CAGA,IAAI,QAA0C;EAC1C,OAAO,KAAK;CAChB;CAGA,IAAI,aAAqB;EACrB,OAAO,KAAK,QAAQ;CACxB;CAGA,MAAM,KACF,GACA,IAAoC,CAAC,GACL;EAChC,OAAO,KAAK,QAAQ,GAAO,CAAO;CACtC;CAGA,MAAM,QACF,GACA,IAAoC,CAAC,GACL;EAChC,IAAM,IAAQ,IAAI,EAAW,GACvB,IAAO,OAAO,KAAU,WAAW,IAAQ,MAC3C,IAAW,MAAM,EAAU,CAAK;EACtC,EAAM,MAAM,MAAM;EAClB,IAAM,IAAS,KAAK,YAAY,CAAQ;EACxC,EAAM,MAAM,YAAY;EACxB,IAAM,IAAU,MAAM,KAAK,SAAS,IAAI,GAAG,KAAK,SAAS,YAAY,EAAO,CAAC;EAC7E,EAAM,MAAM,WAAW;EACvB,IAAM,IAAkB,KAAK,SAAS,YAAY;EAClD,IAAI,MAAoB,KAAA,GACpB,MAAU,MAAM,kCAAkC;EAEtD,IAAM,IAAM,EAAQ;EACpB,IAAI,MAAQ,KAAA,GACR,MAAU,MACN,2BAA2B,EAAgB,4BAC/C;EAEJ,IAAM,IAAY,KAAK,aAAa,EAAI,IAAoB,GAEtD,EAAE,YAAS,cAAW,EAAK,GAAW,EAAQ,QAAQ,IAAI,GAC1D,IAAoC,CAAC;EAC3C,KAAK,IAAI,IAAI,GAAG,IAAI,EAAQ,QAAQ,KAAK;GACrC,IAAM,IAAK,EAAQ,IACb,IAAY,KAAK,QAAQ,MAAO,SAAS;GAC/C,EAAc,KAAK;IACf,SAAS;IACT;IACA,aAAa,EAAO;IACpB,KAAK;IACL,MAAM;IACN,MAAM,EAAO;GACjB,CAAC;EACL;EACA,IAAI,EAAc,WAAW,GACzB,MAAU,MAAM,sDAAsD;EAG1E,IAAM,IAAM,EAAc,IACpB,IAA+B;GACjC,SAAS,EAAI;GACb,WAAW,EAAI;GACf,YAAY,EAAI;GAChB,KAAK,EAAI;GACT,MAAM,EAAI;GACV,MAAM,EAAI;GACV,OAAO;GACP;EACJ,GAEM,IAAkC,CAAC,EAAS,QAAQ,EAAS,KAAK,GAClE,IAAQ,IAAI,EAAM,CAAS;EAEjC,OADA,EAAM,MAAM,aAAa,GAClB,CACH,IAAI,EACA,GACA,GACA,KAAK,QACL,GACA,GACA,GACA,EAAM,MAAM,CAChB,CACJ;CACJ;CAEA,YAAoB,GAA6B;EAC7C,IAAM,CAAC,GAAI,KAAM,KAAK,YAChB,IAAU,EAAO,GAAO,GAAI,CAAE;EAGpC,OAAO,EADK,EADO,EAAU,GAAS,KAAK,OAAO,KAAK,IACrC,GAAY,EAAQ,OAAO,EAAQ,QAAQ,CACtC,GAAK;GAAC;GAAG;GAAG,EAAQ;GAAQ,EAAQ;EAAK,CAAC;CACrE;CAEA,aAAqB,GAAiC;EAClD,OAAO,KAAK,gBAAgB,EAAQ,CAAG,IAAI,IAAI,aAAa,CAAG;CACnE;AACJ"}
+{"version":3,"file":"classifier.js","names":[],"sources":["../../../src/vision/tasks/classifier.ts"],"sourcesContent":["/**\n * Image classification task using ONNX Runtime Web.\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { classificationNumClasses, resolveInputSize } from \"../core/graph\";\nimport { modelNames } from \"../core/metadata\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { softmax, topK } from \"../postprocess/classification\";\nimport { normalize, resize, toCHW, toFloat32Tensor } from \"../preprocess/image\";\nimport { ClassificationResults, Probs } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type ClassProbability, type ClassificationResult, type RGBImage } from \"../types\";\n\nconst IMAGENET_MEAN: readonly [number, number, number] = [0.485, 0.456, 0.406];\nconst IMAGENET_STD: readonly [number, number, number] = [0.229, 0.224, 0.225];\n\nexport interface ClassifierOptions extends OrtSessionOptions {\n    /**\n     * Class label spec — see {@link resolveLabels}.\n     *\n     * Optional: when omitted, the names the export baked into the model are used\n     * (Ultralytics writes them as `names` in the metadata map). Only when the\n     * model carries none does this fall back to generated `class_<id>` labels.\n     * Passing a spec always wins, for a model whose names are wrong or absent.\n     */\n    readonly labels?: LabelSpec;\n    /**\n     * Number of classes the model can predict.\n     *\n     * Optional: inferred from the classification head's declared output shape\n     * `(B, nc)`. Pass it to validate that the supplied labels match the model.\n     */\n    readonly numClasses?: number;\n    /**\n     * Model input `[width, height]` in pixels.\n     *\n     * Only used when the model's graph leaves its spatial axes dynamic: a graph\n     * that declares a static size always wins, since that is the only shape ONNX\n     * Runtime will accept. Defaults to `[224, 224]`.\n     */\n    readonly inputSize?: readonly [number, number];\n    /** Per-channel RGB mean used for normalization. Defaults to ImageNet. */\n    readonly mean?: readonly [number, number, number];\n    /** Per-channel RGB standard deviation. Defaults to ImageNet. */\n    readonly std?: readonly [number, number, number];\n    /**\n     * If `true` (default), apply softmax to the raw model output. Set to\n     * `false` for models whose final layer already produces a probability\n     * distribution.\n     */\n    readonly applySoftmax?: boolean;\n}\n\nexport interface ClassifierPredictOptions {\n    /**\n     * If set, the per-class probability list in `results[0].result.probabilities`\n     * is truncated to the top-K entries. The bulk `probs` view always exposes\n     * the full vector.\n     */\n    readonly topK?: number;\n}\n\n/**\n * Image classifier wrapping an ONNX model with ImageNet-style preprocessing.\n *\n * `predict()` returns `Promise<ClassificationResults[]>` (length 1 for a\n * single image), mirroring Ultralytics' API. The envelope exposes a `probs`\n * collection (`top1`, `top1conf`, `top5`, `top5conf`, `data`) plus the\n * legacy per-class probability list with names already resolved.\n *\n * Defaults: 224×224 RGB input, `float32` normalized with ImageNet mean/std,\n * NCHW layout, batch size 1, softmax applied to the raw output.\n *\n * @example\n * ```typescript\n * const clf = await Classifier.create(\"/models/resnet50.onnx\", {\n *   labels: [\"tench\", \"goldfish\", ...]  // 1000 ImageNet labels\n * });\n * const r = (await clf.predict(\"/images/dog.jpg\"))[0];\n * console.log(r.cls, r.conf, r.name);\n * console.log(r.probs.top5, r.probs.top5conf);\n * ```\n */\nexport class Classifier extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _mean: readonly [number, number, number],\n        private readonly _std: readonly [number, number, number],\n        private readonly _applySoftmax: boolean,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: ClassifierOptions = {}): Promise<Classifier> {\n        const session = await OrtSession.create(model, options);\n        const numClasses =\n            options.numClasses ?? classificationNumClasses(session.outputShape) ?? undefined;\n        const labels = resolveLabels(options.labels ?? modelNames(session.metadata), {\n            numClasses,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Classifier(\n            session,\n            labels,\n            names,\n            resolveInputSize({\n                graphShape: session.inputShape,\n                requested: options.inputSize,\n                fallback: [224, 224],\n            }),\n            options.mean ?? IMAGENET_MEAN,\n            options.std ?? IMAGENET_STD,\n            options.applySoftmax ?? true,\n        );\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /**\n     * The `[width, height]` this task preprocesses to.\n     *\n     * Resolved at creation time from the model's graph when it declares a static\n     * input, so reading it back tells you the resolution inference really runs at\n     * — not merely what was requested.\n     */\n    get inputSize(): readonly [number, number] {\n        return this._inputSize;\n    }\n\n    /** Number of classes the model can predict. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /** Alias for {@link predict} (parity with PyTorch `nn.Module.__call__`). */\n    async call(\n        image: ImageInput,\n        options: ClassifierPredictOptions = {},\n    ): Promise<ClassificationResults[]> {\n        return this.predict(image, options);\n    }\n\n    /** Run classification on a single image. */\n    async predict(\n        image: ImageInput,\n        options: ClassifierPredictOptions = {},\n    ): Promise<ClassificationResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const tensor = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Classifier model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(\n                `Classifier model output ${firstOutputName} missing from run() result.`,\n            );\n        }\n        const fullProbs = this._postprocess(raw.data as Float32Array);\n\n        const { indices, values } = topK(fullProbs, options.topK ?? null);\n        const probabilities: ClassProbability[] = [];\n        for (let i = 0; i < indices.length; i++) {\n            const id = indices[i] as number;\n            const className = this._labels[id] ?? `class_${id}`;\n            probabilities.push({\n                classId: id,\n                className,\n                probability: values[i] as number,\n                cls: id,\n                name: className,\n                conf: values[i] as number,\n            });\n        }\n        if (probabilities.length === 0) {\n            throw new Error(\"Classifier produced no probabilities (empty output).\");\n        }\n\n        const top = probabilities[0] as ClassProbability;\n        const result: ClassificationResult = {\n            classId: top.classId,\n            className: top.className,\n            confidence: top.probability,\n            cls: top.classId,\n            name: top.className,\n            conf: top.probability,\n            image: original,\n            probabilities,\n        };\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const probs = new Probs(fullProbs);\n        timer.stage(\"postprocess\");\n        return [\n            new ClassificationResults(\n                probs,\n                result,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): ort.Tensor {\n        const [tw, th] = this._inputSize;\n        const resized = resize(image, tw, th);\n        const normalized = normalize(resized, this._mean, this._std);\n        const chw = toCHW(normalized, resized.width, resized.height, 3);\n        return toFloat32Tensor(chw, [1, 3, resized.height, resized.width]);\n    }\n\n    private _postprocess(raw: Float32Array): Float32Array {\n        return this._applySoftmax ? softmax(raw) : new Float32Array(raw);\n    }\n}\n"],"mappings":";;;;;;;;;;;AAkBA,IAAM,IAAmD;CAAC;CAAO;CAAO;AAAK,GACvE,IAAkD;CAAC;CAAO;CAAO;AAAK,GAqE/D,IAAb,MAAa,UAAmB,EAAW;CAGlB;CACA;CACA;CACA;CACA;CACA;CAPrB,YACI,GACA,GACA,GACA,GACA,GACA,GACA,GACF;EADmB,AAEjB,MAAM,CAAO,GAPI,KAAA,UAAA,GACA,KAAA,SAAA,GACA,KAAA,aAAA,GACA,KAAA,QAAA,GACA,KAAA,OAAA,GACA,KAAA,gBAAA;CAGrB;CAGA,aAAa,OAAO,GAAoB,IAA6B,CAAC,GAAwB;EAC1F,IAAM,IAAU,MAAM,EAAW,OAAO,GAAO,CAAO,GAChD,IACF,EAAQ,cAAc,EAAyB,EAAQ,WAAW,KAAK,KAAA,GACrE,IAAS,EAAc,EAAQ,UAAU,EAAW,EAAQ,QAAQ,GAAG,EACzE,cACJ,CAAC,GACK,IAAgC,CAAC;EACvC,KAAK,IAAI,IAAI,GAAG,IAAI,EAAO,QAAQ,KAC/B,EAAM,KAAK,EAAO;EAEtB,OAAO,IAAI,EACP,GACA,GACA,GACA,EAAiB;GACb,YAAY,EAAQ;GACpB,WAAW,EAAQ;GACnB,UAAU,CAAC,KAAK,GAAG;EACvB,CAAC,GACD,EAAQ,QAAQ,GAChB,EAAQ,OAAO,GACf,EAAQ,gBAAgB,EAC5B;CACJ;CAGA,IAAI,SAA4B;EAC5B,OAAO,KAAK;CAChB;CAGA,IAAI,QAA0C;EAC1C,OAAO,KAAK;CAChB;CASA,IAAI,YAAuC;EACvC,OAAO,KAAK;CAChB;CAGA,IAAI,aAAqB;EACrB,OAAO,KAAK,QAAQ;CACxB;CAGA,MAAM,KACF,GACA,IAAoC,CAAC,GACL;EAChC,OAAO,KAAK,QAAQ,GAAO,CAAO;CACtC;CAGA,MAAM,QACF,GACA,IAAoC,CAAC,GACL;EAChC,IAAM,IAAQ,IAAI,EAAW,GACvB,IAAO,OAAO,KAAU,WAAW,IAAQ,MAC3C,IAAW,MAAM,EAAU,CAAK;EACtC,EAAM,MAAM,MAAM;EAClB,IAAM,IAAS,KAAK,YAAY,CAAQ;EACxC,EAAM,MAAM,YAAY;EACxB,IAAM,IAAU,MAAM,KAAK,SAAS,IAAI,GAAG,KAAK,SAAS,YAAY,EAAO,CAAC;EAC7E,EAAM,MAAM,WAAW;EACvB,IAAM,IAAkB,KAAK,SAAS,YAAY;EAClD,IAAI,MAAoB,KAAA,GACpB,MAAU,MAAM,kCAAkC;EAEtD,IAAM,IAAM,EAAQ;EACpB,IAAI,MAAQ,KAAA,GACR,MAAU,MACN,2BAA2B,EAAgB,4BAC/C;EAEJ,IAAM,IAAY,KAAK,aAAa,EAAI,IAAoB,GAEtD,EAAE,YAAS,cAAW,EAAK,GAAW,EAAQ,QAAQ,IAAI,GAC1D,IAAoC,CAAC;EAC3C,KAAK,IAAI,IAAI,GAAG,IAAI,EAAQ,QAAQ,KAAK;GACrC,IAAM,IAAK,EAAQ,IACb,IAAY,KAAK,QAAQ,MAAO,SAAS;GAC/C,EAAc,KAAK;IACf,SAAS;IACT;IACA,aAAa,EAAO;IACpB,KAAK;IACL,MAAM;IACN,MAAM,EAAO;GACjB,CAAC;EACL;EACA,IAAI,EAAc,WAAW,GACzB,MAAU,MAAM,sDAAsD;EAG1E,IAAM,IAAM,EAAc,IACpB,IAA+B;GACjC,SAAS,EAAI;GACb,WAAW,EAAI;GACf,YAAY,EAAI;GAChB,KAAK,EAAI;GACT,MAAM,EAAI;GACV,MAAM,EAAI;GACV,OAAO;GACP;EACJ,GAEM,IAAkC,CAAC,EAAS,QAAQ,EAAS,KAAK,GAClE,IAAQ,IAAI,EAAM,CAAS;EAEjC,OADA,EAAM,MAAM,aAAa,GAClB,CACH,IAAI,EACA,GACA,GACA,KAAK,QACL,GACA,GACA,GACA,EAAM,MAAM,CAChB,CACJ;CACJ;CAEA,YAAoB,GAA6B;EAC7C,IAAM,CAAC,GAAI,KAAM,KAAK,YAChB,IAAU,EAAO,GAAO,GAAI,CAAE;EAGpC,OAAO,EADK,EADO,EAAU,GAAS,KAAK,OAAO,KAAK,IACrC,GAAY,EAAQ,OAAO,EAAQ,QAAQ,CACtC,GAAK;GAAC;GAAG;GAAG,EAAQ;GAAQ,EAAQ;EAAK,CAAC;CACrE;CAEA,aAAqB,GAAiC;EAClD,OAAO,KAAK,gBAAgB,EAAQ,CAAG,IAAI,IAAI,aAAa,CAAG;CACnE;AACJ"}

package/dist/vision/tasks/detector.cjs

@@ -1,2 +1,2 @@
-const e=require("../types.cjs"),t=require("../core/timing.cjs"),n=require("../results.cjs"),r=require("../labels.cjs"),i=require("../core/session.cjs"),a=require("../io/image.cjs"),o=require("../preprocess/image.cjs"),s=require("../postprocess/detection.cjs"),c=require("./base.cjs");var l=class l extends c.VisionTask{_head;_labels;_names;_inputSize;_confThreshold;_iouThreshold;_maxDetections;constructor(e,t,n,r,i,a,o,s){super(e),this._head=t,this._labels=n,this._names=r,this._inputSize=i,this._confThreshold=a,this._iouThreshold=o,this._maxDetections=s}static async create(e,t={}){let n=t.head??`yolo`;if(n!==`yolo`)throw Error(`Unsupported detector head '${n}'. Supported: 'yolo'.`);let a=await i.OrtSession.create(e,t),o=r.resolveLabels(t.labels??`coco`,{numClasses:t.numClasses}),s={};for(let e=0;e<o.length;e++)s[e]=o[e];return new l(a,n,o,s,t.inputSize??[640,640],t.confThreshold??.25,t.iouThreshold??.45,t.maxDetections??300)}get head(){return this._head}get labels(){return this._labels}get names(){return this._names}get numClasses(){return this._labels.length}async call(e,t={}){return this.predict(e,t)}async predict(e,r={}){let i=new t.SpeedTimer,o=typeof e==`string`?e:null,c=await a.loadImage(e);i.stage(`load`);let{tensor:l,scale:u,padLeft:d,padTop:f}=this._preprocess(c);i.stage(`preprocess`);let p=await this._session.run({[this._session.inputName]:l});i.stage(`inference`);let m=this._session.outputNames[0];if(m===void 0)throw Error(`Detector model has no outputs.`);let h=p[m];if(h===void 0)throw Error(`Detector model output ${m} missing from run() result.`);let g=s.decodeYolo(h.data,h.dims,{originalWidth:c.width,originalHeight:c.height,padLeft:d,padTop:f,scale:u,confThreshold:r.confThreshold??this._confThreshold,iouThreshold:r.iouThreshold??this._iouThreshold,maxDetections:this._maxDetections}),_=(r.classes===void 0?g:(()=>{let e=new Set(r.classes);return g.filter(t=>e.has(t.classId))})()).map(e=>this._buildResult(c,e.bbox,e.classId,e.confidence)),v=[c.height,c.width],y=this._buildBoxes(_,v);return i.stage(`postprocess`),[new n.DetectionResults(y,_,this._names,c,v,o,i.speed())]}_preprocess(e){let[t,n]=this._inputSize,r=o.letterbox(e,t,n);return{tensor:o.toFloat32Tensor(o.toCHW(o.toFloat32(r.image),r.image.width,r.image.height,3),[1,3,r.image.height,r.image.width]),scale:r.scale,padLeft:r.padLeft,padTop:r.padTop}}_buildResult(t,n,r,i){let[a,o,s,c]=n.asIntXyxy(),l=Math.max(0,a),u=Math.max(0,o),d=Math.min(t.width,s),f=Math.min(t.height,c),p;if(d>l&&f>u){let n=d-l,r=f-u,i=new Uint8Array(n*r*3);for(let e=0;e<r;e++){let r=((u+e)*t.width+l)*3;i.set(t.data.subarray(r,r+n*3),e*n*3)}p=new e.RGBImage(i,n,r)}else p=new e.RGBImage(new Uint8Array,0,0);let m=this._names[r]??`class_${r}`;return{classId:r,className:m,confidence:i,bbox:n,cls:r,name:m,conf:i,box:n,croppedImage:p}}_buildBoxes(e,t){let r=e.length,i=new Float32Array(r*4),a=new Int32Array(r),o=new Float32Array(r);for(let t=0;t<r;t++){let n=e[t];i[t*4]=n.bbox.x1,i[t*4+1]=n.bbox.y1,i[t*4+2]=n.bbox.x2,i[t*4+3]=n.bbox.y2,a[t]=n.classId,o[t]=n.confidence}return new n.Boxes(i,a,o,t)}};exports.Detector=l;
+const e=require("../types.cjs"),t=require("../core/timing.cjs"),n=require("../results.cjs"),r=require("../labels.cjs"),i=require("../core/graph.cjs"),a=require("../core/metadata.cjs"),o=require("../core/session.cjs"),s=require("../io/image.cjs"),c=require("../preprocess/image.cjs"),l=require("../postprocess/detection.cjs"),u=require("./base.cjs");var d=class d extends u.VisionTask{_head;_labels;_names;_inputSize;_confThreshold;_iouThreshold;_maxDetections;constructor(e,t,n,r,i,a,o,s){super(e),this._head=t,this._labels=n,this._names=r,this._inputSize=i,this._confThreshold=a,this._iouThreshold=o,this._maxDetections=s}static async create(e,t={}){let n=t.head??`yolo`;if(n!==`yolo`)throw Error(`Unsupported detector head '${n}'. Supported: 'yolo'.`);let s=await o.OrtSession.create(e,t),c=r.resolveLabels(t.labels??a.modelNames(s.metadata)??`coco`,{numClasses:t.numClasses??i.detectionNumClasses(s.outputShape)??void 0}),l={};for(let e=0;e<c.length;e++)l[e]=c[e];return new d(s,n,c,l,i.resolveInputSize({graphShape:s.inputShape,requested:t.inputSize,fallback:[640,640]}),t.confThreshold??.25,t.iouThreshold??.45,t.maxDetections??300)}get head(){return this._head}get labels(){return this._labels}get names(){return this._names}get inputSize(){return this._inputSize}get numClasses(){return this._labels.length}async call(e,t={}){return this.predict(e,t)}async predict(e,r={}){let i=new t.SpeedTimer,a=typeof e==`string`?e:null,o=await s.loadImage(e);i.stage(`load`);let{tensor:c,scale:u,padLeft:d,padTop:f}=this._preprocess(o);i.stage(`preprocess`);let p=await this._session.run({[this._session.inputName]:c});i.stage(`inference`);let m=this._session.outputNames[0];if(m===void 0)throw Error(`Detector model has no outputs.`);let h=p[m];if(h===void 0)throw Error(`Detector model output ${m} missing from run() result.`);let g=l.decodeYolo(h.data,h.dims,{originalWidth:o.width,originalHeight:o.height,padLeft:d,padTop:f,scale:u,confThreshold:r.confThreshold??this._confThreshold,iouThreshold:r.iouThreshold??this._iouThreshold,maxDetections:this._maxDetections}),_=(r.classes===void 0?g:(()=>{let e=new Set(r.classes);return g.filter(t=>e.has(t.classId))})()).map(e=>this._buildResult(o,e.bbox,e.classId,e.confidence)),v=[o.height,o.width],y=this._buildBoxes(_,v);return i.stage(`postprocess`),[new n.DetectionResults(y,_,this._names,o,v,a,i.speed())]}_preprocess(e){let[t,n]=this._inputSize,r=c.letterbox(e,t,n);return{tensor:c.toFloat32Tensor(c.toCHW(c.toFloat32(r.image),r.image.width,r.image.height,3),[1,3,r.image.height,r.image.width]),scale:r.scale,padLeft:r.padLeft,padTop:r.padTop}}_buildResult(t,n,r,i){let[a,o,s,c]=n.asIntXyxy(),l=Math.max(0,a),u=Math.max(0,o),d=Math.min(t.width,s),f=Math.min(t.height,c),p;if(d>l&&f>u){let n=d-l,r=f-u,i=new Uint8Array(n*r*3);for(let e=0;e<r;e++){let r=((u+e)*t.width+l)*3;i.set(t.data.subarray(r,r+n*3),e*n*3)}p=new e.RGBImage(i,n,r)}else p=new e.RGBImage(new Uint8Array,0,0);let m=this._names[r]??`class_${r}`;return{classId:r,className:m,confidence:i,bbox:n,cls:r,name:m,conf:i,box:n,croppedImage:p}}_buildBoxes(e,t){let r=e.length,i=new Float32Array(r*4),a=new Int32Array(r),o=new Float32Array(r);for(let t=0;t<r;t++){let n=e[t];i[t*4]=n.bbox.x1,i[t*4+1]=n.bbox.y1,i[t*4+2]=n.bbox.x2,i[t*4+3]=n.bbox.y2,a[t]=n.classId,o[t]=n.confidence}return new n.Boxes(i,a,o,t)}};exports.Detector=d;
 //# sourceMappingURL=detector.cjs.map

package/dist/vision/tasks/detector.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"detector.cjs","names":[],"sources":["../../../src/vision/tasks/detector.ts"],"sourcesContent":["/**\n * Object detection task using anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYolo } from \"../postprocess/detection\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, DetectionResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type DetectionResult, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the model's detection head.\n *\n * - `\"yolo\"`: anchor-free YOLO head with output shape `[1, 4 + nc, N]` —\n *   covers YOLOv8, v9, v10, v11, v12, v26 detect exports.\n *\n * The SDK does **not** auto-detect the head from the model — the caller is\n * responsible for picking a head that matches their export. Future families\n * (v5/v6/v7 with `[1, N, 5+nc]`) will be added as new literal members.\n */\nexport type DetectorHead = \"yolo\";\n\nexport interface DetectorOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the detection head. Default `\"yolo\"` covers\n     * YOLOv8/v9/v10/v11/v12/v26.\n     */\n    readonly head?: DetectorHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /** Model input `[width, height]` for letterboxing. Defaults to `[640, 640]`. */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of detections per image. */\n    readonly maxDetections?: number;\n}\n\nexport interface DetectorPredictOptions {\n    /** Override the default confidence threshold. */\n    readonly confThreshold?: number;\n    /** Override the default IoU threshold. */\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only detections whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Object detector for anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n *\n * `predict()` returns `Promise<DetectionResults[]>` (length 1 for a single\n * image), mirroring Ultralytics' `YOLO(\"img.jpg\")`. Iterate the envelope for\n * per-instance dataclasses, or use the bulk `boxes` view (`.xyxy`, `.xywh`,\n * `.xyxyn`, `.xywhn`, `.cls`, `.conf`).\n *\n * @example\n * ```typescript\n * const det = await Detector.create(\"/models/yolov8n.onnx\");\n * const results = await det.predict(\"/images/street.jpg\");\n * const r = results[0];\n * console.log(r.boxes.xyxy, r.boxes.cls, r.boxes.conf, r.names);\n * for (const d of r) {\n *   console.log(d.cls, d.conf, d.box.xyxy);\n * }\n * ```\n */\nexport class Detector extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: DetectorHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: DetectorOptions = {}): Promise<Detector> {\n        const head: DetectorHead = options.head ?? \"yolo\";\n        if (head !== \"yolo\") {\n            throw new Error(`Unsupported detector head '${head}'. Supported: 'yolo'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? \"coco\", {\n            numClasses: options.numClasses,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Detector(\n            session,\n            head,\n            labels,\n            names,\n            options.inputSize ?? [640, 640],\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): DetectorHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /**\n     * Alias for {@link predict} — call the detector like a torch `nn.Module`.\n     *\n     * Use as `det.call(img)` since JavaScript class instances are not callable;\n     * for direct invocation, prefer `det.predict(img)`. The full\n     * {@link DetectorPredictOptions} (including `classes`) is supported.\n     */\n    async call(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        return this.predict(image, options);\n    }\n\n    /**\n     * Run detection on a single image.\n     *\n     * The returned envelope carries a {@link Speed} breakdown in `speed`,\n     * mirroring Ultralytics' `results[0].speed`.\n     */\n    async predict(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Detector model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(`Detector model output ${firstOutputName} missing from run() result.`);\n        }\n\n        const decodedAll = decodeYolo(raw.data as Float32Array, raw.dims, {\n            originalWidth: original.width,\n            originalHeight: original.height,\n            padLeft,\n            padTop,\n            scale,\n            confThreshold: options.confThreshold ?? this._confThreshold,\n            iouThreshold: options.iouThreshold ?? this._iouThreshold,\n            maxDetections: this._maxDetections,\n        });\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new DetectionResults(\n                boxes,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n    ): DetectionResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let cropped: RGBImage;\n        if (cx2 > cx1 && cy2 > cy1) {\n            const cw = cx2 - cx1;\n            const ch = cy2 - cy1;\n            const out = new Uint8Array(cw * ch * 3);\n            for (let row = 0; row < ch; row++) {\n                const srcOffset = ((cy1 + row) * original.width + cx1) * 3;\n                out.set(original.data.subarray(srcOffset, srcOffset + cw * 3), row * cw * 3);\n            }\n            cropped = new RGBImage(out, cw, ch);\n        } else {\n            cropped = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            croppedImage: cropped,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly DetectionResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as DetectionResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n}\n"],"mappings":"4RA+EA,IAAa,EAAb,MAAa,UAAiB,EAAA,UAAW,CAGhB,MACA,QACA,OACA,WACA,eACA,cACA,eARrB,YACI,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACF,CACE,MAAM,CAAO,EARI,KAAA,MAAA,EACA,KAAA,QAAA,EACA,KAAA,OAAA,EACA,KAAA,WAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,eAAA,CAGrB,CAGA,aAAa,OAAO,EAAoB,EAA2B,CAAC,EAAsB,CACtF,IAAM,EAAqB,EAAQ,MAAQ,OAC3C,GAAI,IAAS,OACT,MAAU,MAAM,8BAA8B,EAAK,sBAAsB,EAE7E,IAAM,EAAU,MAAM,EAAA,WAAW,OAAO,EAAO,CAAO,EAChD,EAAS,EAAA,cAAc,EAAQ,QAAU,OAAQ,CACnD,WAAY,EAAQ,UACxB,CAAC,EACK,EAAgC,CAAC,EACvC,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAM,GAAK,EAAO,GAEtB,OAAO,IAAI,EACP,EACA,EACA,EACA,EACA,EAAQ,WAAa,CAAC,IAAK,GAAG,EAC9B,EAAQ,eAAiB,IACzB,EAAQ,cAAgB,IACxB,EAAQ,eAAiB,GAC7B,CACJ,CAGA,IAAI,MAAqB,CACrB,OAAO,KAAK,KAChB,CAGA,IAAI,QAA4B,CAC5B,OAAO,KAAK,OAChB,CAGA,IAAI,OAA0C,CAC1C,OAAO,KAAK,MAChB,CAGA,IAAI,YAAqB,CACrB,OAAO,KAAK,QAAQ,MACxB,CASA,MAAM,KACF,EACA,EAAkC,CAAC,EACR,CAC3B,OAAO,KAAK,QAAQ,EAAO,CAAO,CACtC,CAQA,MAAM,QACF,EACA,EAAkC,CAAC,EACR,CAC3B,IAAM,EAAQ,IAAI,EAAA,WACZ,EAAO,OAAO,GAAU,SAAW,EAAQ,KAC3C,EAAW,MAAM,EAAA,UAAU,CAAK,EACtC,EAAM,MAAM,MAAM,EAClB,GAAM,CAAE,SAAQ,QAAO,UAAS,UAAW,KAAK,YAAY,CAAQ,EACpE,EAAM,MAAM,YAAY,EACxB,IAAM,EAAU,MAAM,KAAK,SAAS,IAAI,EAAG,KAAK,SAAS,WAAY,CAAO,CAAC,EAC7E,EAAM,MAAM,WAAW,EAEvB,IAAM,EAAkB,KAAK,SAAS,YAAY,GAClD,GAAI,IAAoB,IAAA,GACpB,MAAU,MAAM,gCAAgC,EAEpD,IAAM,EAAM,EAAQ,GACpB,GAAI,IAAQ,IAAA,GACR,MAAU,MAAM,yBAAyB,EAAgB,4BAA4B,EAGzF,IAAM,EAAa,EAAA,WAAW,EAAI,KAAsB,EAAI,KAAM,CAC9D,cAAe,EAAS,MACxB,eAAgB,EAAS,OACzB,UACA,SACA,QACA,cAAe,EAAQ,eAAiB,KAAK,eAC7C,aAAc,EAAQ,cAAgB,KAAK,cAC3C,cAAe,KAAK,cACxB,CAAC,EAUK,GAPF,EAAQ,UAAY,IAAA,GAKd,OAJO,CACH,IAAM,EAAU,IAAI,IAAI,EAAQ,OAAO,EACvC,OAAO,EAAW,OAAQ,GAAM,EAAQ,IAAI,EAAE,OAAO,CAAC,CAC1D,EAAA,CAAG,EACH,CAEiB,IAAK,GAC5B,KAAK,aAAa,EAAU,EAAE,KAAM,EAAE,QAAS,EAAE,UAAU,CAC/D,EAEM,EAAkC,CAAC,EAAS,OAAQ,EAAS,KAAK,EAClE,EAAQ,KAAK,YAAY,EAAY,CAAI,EAE/C,OADA,EAAM,MAAM,aAAa,EAClB,CACH,IAAI,EAAA,iBACA,EACA,EACA,KAAK,OACL,EACA,EACA,EACA,EAAM,MAAM,CAChB,CACJ,CACJ,CAEA,YAAoB,EAKlB,CACE,GAAM,CAAC,EAAI,GAAM,KAAK,WAChB,EAAK,EAAA,UAAU,EAAO,EAAI,CAAE,EAGlC,MAAO,CACH,OAAQ,EAAA,gBAFA,EAAA,MADA,EAAA,UAAU,EAAG,KACP,EAAK,EAAG,MAAM,MAAO,EAAG,MAAM,OAAQ,CAE5B,EAAK,CAAC,EAAG,EAAG,EAAG,MAAM,OAAQ,EAAG,MAAM,KAAK,CAAC,EACpE,MAAO,EAAG,MACV,QAAS,EAAG,QACZ,OAAQ,EAAG,MACf,CACJ,CAEA,aACI,EACA,EACA,EACA,EACe,CACf,GAAM,CAAC,EAAI,EAAI,EAAI,GAAM,EAAK,UAAU,EAClC,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAS,MAAO,CAAE,EACjC,EAAM,KAAK,IAAI,EAAS,OAAQ,CAAE,EAEpC,EACJ,GAAI,EAAM,GAAO,EAAM,EAAK,CACxB,IAAM,EAAK,EAAM,EACX,EAAK,EAAM,EACX,EAAM,IAAI,WAAW,EAAK,EAAK,CAAC,EACtC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAAO,CAC/B,IAAM,IAAc,EAAM,GAAO,EAAS,MAAQ,GAAO,EACzD,EAAI,IAAI,EAAS,KAAK,SAAS,EAAW,EAAY,EAAK,CAAC,EAAG,EAAM,EAAK,CAAC,CAC/E,CACA,EAAU,IAAI,EAAA,SAAS,EAAK,EAAI,CAAE,CACtC,KACI,GAAU,IAAI,EAAA,SAAS,IAAI,WAAe,EAAG,CAAC,EAGlD,IAAM,EAAY,KAAK,OAAO,IAAY,SAAS,IAEnD,MAAO,CACH,UACA,YACA,aACA,OACA,IAAK,EACL,KAAM,EACN,KAAM,EACN,IAAK,EACL,aAAc,CAClB,CACJ,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAI,EAAW,OACf,EAAO,IAAI,aAAa,EAAI,CAAC,EAC7B,EAAM,IAAI,WAAW,CAAC,EACtB,EAAO,IAAI,aAAa,CAAC,EAC/B,IAAK,IAAI,EAAI,EAAG,EAAI,EAAG,IAAK,CACxB,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAI,GAAK,EAAE,QACX,EAAK,GAAK,EAAE,UAChB,CACA,OAAO,IAAI,EAAA,MAAM,EAAM,EAAK,EAAM,CAAS,CAC/C,CACJ"}
+{"version":3,"file":"detector.cjs","names":[],"sources":["../../../src/vision/tasks/detector.ts"],"sourcesContent":["/**\n * Object detection task using anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\n\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { detectionNumClasses, resolveInputSize } from \"../core/graph\";\nimport { modelNames } from \"../core/metadata\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYolo } from \"../postprocess/detection\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, DetectionResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type DetectionResult, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the model's detection head.\n *\n * - `\"yolo\"`: anchor-free YOLO head with output shape `[1, 4 + nc, N]` —\n *   covers YOLOv8, v9, v10, v11, v12, v26 detect exports.\n *\n * The SDK does **not** auto-detect the head from the model — the caller is\n * responsible for picking a head that matches their export. Future families\n * (v5/v6/v7 with `[1, N, 5+nc]`) will be added as new literal members.\n */\nexport type DetectorHead = \"yolo\";\n\nexport interface DetectorOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the detection head. Default `\"yolo\"` covers\n     * YOLOv8/v9/v10/v11/v12/v26.\n     */\n    readonly head?: DetectorHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /**\n     * Model input `[width, height]` in pixels for letterboxing.\n     *\n     * Only used when the model's graph leaves its spatial axes dynamic: a graph\n     * that declares a static size always wins, since that is the only shape ONNX\n     * Runtime will accept. Defaults to `[640, 640]`.\n     */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of detections per image. */\n    readonly maxDetections?: number;\n}\n\nexport interface DetectorPredictOptions {\n    /** Override the default confidence threshold. */\n    readonly confThreshold?: number;\n    /** Override the default IoU threshold. */\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only detections whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Object detector for anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n *\n * `predict()` returns `Promise<DetectionResults[]>` (length 1 for a single\n * image), mirroring Ultralytics' `YOLO(\"img.jpg\")`. Iterate the envelope for\n * per-instance dataclasses, or use the bulk `boxes` view (`.xyxy`, `.xywh`,\n * `.xyxyn`, `.xywhn`, `.cls`, `.conf`).\n *\n * @example\n * ```typescript\n * const det = await Detector.create(\"/models/yolov8n.onnx\");\n * const results = await det.predict(\"/images/street.jpg\");\n * const r = results[0];\n * console.log(r.boxes.xyxy, r.boxes.cls, r.boxes.conf, r.names);\n * for (const d of r) {\n *   console.log(d.cls, d.conf, d.box.xyxy);\n * }\n * ```\n */\nexport class Detector extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: DetectorHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: DetectorOptions = {}): Promise<Detector> {\n        const head: DetectorHead = options.head ?? \"yolo\";\n        if (head !== \"yolo\") {\n            throw new Error(`Unsupported detector head '${head}'. Supported: 'yolo'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? modelNames(session.metadata) ?? \"coco\", {\n            numClasses: options.numClasses ?? detectionNumClasses(session.outputShape) ?? undefined,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Detector(\n            session,\n            head,\n            labels,\n            names,\n            resolveInputSize({\n                graphShape: session.inputShape,\n                requested: options.inputSize,\n                fallback: [640, 640],\n            }),\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): DetectorHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /**\n     * The `[width, height]` this task preprocesses to.\n     *\n     * Resolved at creation time from the model's graph when it declares a static\n     * input, so reading it back tells you the resolution inference really runs at\n     * — not merely what was requested.\n     */\n    get inputSize(): readonly [number, number] {\n        return this._inputSize;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /**\n     * Alias for {@link predict} — call the detector like a torch `nn.Module`.\n     *\n     * Use as `det.call(img)` since JavaScript class instances are not callable;\n     * for direct invocation, prefer `det.predict(img)`. The full\n     * {@link DetectorPredictOptions} (including `classes`) is supported.\n     */\n    async call(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        return this.predict(image, options);\n    }\n\n    /**\n     * Run detection on a single image.\n     *\n     * The returned envelope carries a {@link Speed} breakdown in `speed`,\n     * mirroring Ultralytics' `results[0].speed`.\n     */\n    async predict(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Detector model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(`Detector model output ${firstOutputName} missing from run() result.`);\n        }\n\n        const decodedAll = decodeYolo(raw.data as Float32Array, raw.dims, {\n            originalWidth: original.width,\n            originalHeight: original.height,\n            padLeft,\n            padTop,\n            scale,\n            confThreshold: options.confThreshold ?? this._confThreshold,\n            iouThreshold: options.iouThreshold ?? this._iouThreshold,\n            maxDetections: this._maxDetections,\n        });\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new DetectionResults(\n                boxes,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n    ): DetectionResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let cropped: RGBImage;\n        if (cx2 > cx1 && cy2 > cy1) {\n            const cw = cx2 - cx1;\n            const ch = cy2 - cy1;\n            const out = new Uint8Array(cw * ch * 3);\n            for (let row = 0; row < ch; row++) {\n                const srcOffset = ((cy1 + row) * original.width + cx1) * 3;\n                out.set(original.data.subarray(srcOffset, srcOffset + cw * 3), row * cw * 3);\n            }\n            cropped = new RGBImage(out, cw, ch);\n        } else {\n            cropped = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            croppedImage: cropped,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly DetectionResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as DetectionResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n}\n"],"mappings":"6VAwFA,IAAa,EAAb,MAAa,UAAiB,EAAA,UAAW,CAGhB,MACA,QACA,OACA,WACA,eACA,cACA,eARrB,YACI,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACF,CACE,MAAM,CAAO,EARI,KAAA,MAAA,EACA,KAAA,QAAA,EACA,KAAA,OAAA,EACA,KAAA,WAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,eAAA,CAGrB,CAGA,aAAa,OAAO,EAAoB,EAA2B,CAAC,EAAsB,CACtF,IAAM,EAAqB,EAAQ,MAAQ,OAC3C,GAAI,IAAS,OACT,MAAU,MAAM,8BAA8B,EAAK,sBAAsB,EAE7E,IAAM,EAAU,MAAM,EAAA,WAAW,OAAO,EAAO,CAAO,EAChD,EAAS,EAAA,cAAc,EAAQ,QAAU,EAAA,WAAW,EAAQ,QAAQ,GAAK,OAAQ,CACnF,WAAY,EAAQ,YAAc,EAAA,oBAAoB,EAAQ,WAAW,GAAK,IAAA,EAClF,CAAC,EACK,EAAgC,CAAC,EACvC,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAM,GAAK,EAAO,GAEtB,OAAO,IAAI,EACP,EACA,EACA,EACA,EACA,EAAA,iBAAiB,CACb,WAAY,EAAQ,WACpB,UAAW,EAAQ,UACnB,SAAU,CAAC,IAAK,GAAG,CACvB,CAAC,EACD,EAAQ,eAAiB,IACzB,EAAQ,cAAgB,IACxB,EAAQ,eAAiB,GAC7B,CACJ,CAGA,IAAI,MAAqB,CACrB,OAAO,KAAK,KAChB,CAGA,IAAI,QAA4B,CAC5B,OAAO,KAAK,OAChB,CAGA,IAAI,OAA0C,CAC1C,OAAO,KAAK,MAChB,CASA,IAAI,WAAuC,CACvC,OAAO,KAAK,UAChB,CAGA,IAAI,YAAqB,CACrB,OAAO,KAAK,QAAQ,MACxB,CASA,MAAM,KACF,EACA,EAAkC,CAAC,EACR,CAC3B,OAAO,KAAK,QAAQ,EAAO,CAAO,CACtC,CAQA,MAAM,QACF,EACA,EAAkC,CAAC,EACR,CAC3B,IAAM,EAAQ,IAAI,EAAA,WACZ,EAAO,OAAO,GAAU,SAAW,EAAQ,KAC3C,EAAW,MAAM,EAAA,UAAU,CAAK,EACtC,EAAM,MAAM,MAAM,EAClB,GAAM,CAAE,SAAQ,QAAO,UAAS,UAAW,KAAK,YAAY,CAAQ,EACpE,EAAM,MAAM,YAAY,EACxB,IAAM,EAAU,MAAM,KAAK,SAAS,IAAI,EAAG,KAAK,SAAS,WAAY,CAAO,CAAC,EAC7E,EAAM,MAAM,WAAW,EAEvB,IAAM,EAAkB,KAAK,SAAS,YAAY,GAClD,GAAI,IAAoB,IAAA,GACpB,MAAU,MAAM,gCAAgC,EAEpD,IAAM,EAAM,EAAQ,GACpB,GAAI,IAAQ,IAAA,GACR,MAAU,MAAM,yBAAyB,EAAgB,4BAA4B,EAGzF,IAAM,EAAa,EAAA,WAAW,EAAI,KAAsB,EAAI,KAAM,CAC9D,cAAe,EAAS,MACxB,eAAgB,EAAS,OACzB,UACA,SACA,QACA,cAAe,EAAQ,eAAiB,KAAK,eAC7C,aAAc,EAAQ,cAAgB,KAAK,cAC3C,cAAe,KAAK,cACxB,CAAC,EAUK,GAPF,EAAQ,UAAY,IAAA,GAKd,OAJO,CACH,IAAM,EAAU,IAAI,IAAI,EAAQ,OAAO,EACvC,OAAO,EAAW,OAAQ,GAAM,EAAQ,IAAI,EAAE,OAAO,CAAC,CAC1D,EAAA,CAAG,EACH,CAEiB,IAAK,GAC5B,KAAK,aAAa,EAAU,EAAE,KAAM,EAAE,QAAS,EAAE,UAAU,CAC/D,EAEM,EAAkC,CAAC,EAAS,OAAQ,EAAS,KAAK,EAClE,EAAQ,KAAK,YAAY,EAAY,CAAI,EAE/C,OADA,EAAM,MAAM,aAAa,EAClB,CACH,IAAI,EAAA,iBACA,EACA,EACA,KAAK,OACL,EACA,EACA,EACA,EAAM,MAAM,CAChB,CACJ,CACJ,CAEA,YAAoB,EAKlB,CACE,GAAM,CAAC,EAAI,GAAM,KAAK,WAChB,EAAK,EAAA,UAAU,EAAO,EAAI,CAAE,EAGlC,MAAO,CACH,OAAQ,EAAA,gBAFA,EAAA,MADA,EAAA,UAAU,EAAG,KACP,EAAK,EAAG,MAAM,MAAO,EAAG,MAAM,OAAQ,CAE5B,EAAK,CAAC,EAAG,EAAG,EAAG,MAAM,OAAQ,EAAG,MAAM,KAAK,CAAC,EACpE,MAAO,EAAG,MACV,QAAS,EAAG,QACZ,OAAQ,EAAG,MACf,CACJ,CAEA,aACI,EACA,EACA,EACA,EACe,CACf,GAAM,CAAC,EAAI,EAAI,EAAI,GAAM,EAAK,UAAU,EAClC,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAS,MAAO,CAAE,EACjC,EAAM,KAAK,IAAI,EAAS,OAAQ,CAAE,EAEpC,EACJ,GAAI,EAAM,GAAO,EAAM,EAAK,CACxB,IAAM,EAAK,EAAM,EACX,EAAK,EAAM,EACX,EAAM,IAAI,WAAW,EAAK,EAAK,CAAC,EACtC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAAO,CAC/B,IAAM,IAAc,EAAM,GAAO,EAAS,MAAQ,GAAO,EACzD,EAAI,IAAI,EAAS,KAAK,SAAS,EAAW,EAAY,EAAK,CAAC,EAAG,EAAM,EAAK,CAAC,CAC/E,CACA,EAAU,IAAI,EAAA,SAAS,EAAK,EAAI,CAAE,CACtC,KACI,GAAU,IAAI,EAAA,SAAS,IAAI,WAAe,EAAG,CAAC,EAGlD,IAAM,EAAY,KAAK,OAAO,IAAY,SAAS,IAEnD,MAAO,CACH,UACA,YACA,aACA,OACA,IAAK,EACL,KAAM,EACN,KAAM,EACN,IAAK,EACL,aAAc,CAClB,CACJ,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAI,EAAW,OACf,EAAO,IAAI,aAAa,EAAI,CAAC,EAC7B,EAAM,IAAI,WAAW,CAAC,EACtB,EAAO,IAAI,aAAa,CAAC,EAC/B,IAAK,IAAI,EAAI,EAAG,EAAI,EAAG,IAAK,CACxB,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAI,GAAK,EAAE,QACX,EAAK,GAAK,EAAE,UAChB,CACA,OAAO,IAAI,EAAA,MAAM,EAAM,EAAK,EAAM,CAAS,CAC/C,CACJ"}

package/dist/vision/tasks/detector.js

@@ -2,13 +2,15 @@ import { RGBImage as e } from "../types.js";
 import { SpeedTimer as t } from "../core/timing.js";
 import { Boxes as n, DetectionResults as r } from "../results.js";
 import { resolveLabels as i } from "../labels.js";
-import { OrtSession as a } from "../core/session.js";
-import { loadImage as o } from "../io/image.js";
-import { letterbox as s, toCHW as c, toFloat32 as l, toFloat32Tensor as u } from "../preprocess/image.js";
-import { decodeYolo as d } from "../postprocess/detection.js";
-import { VisionTask as f } from "./base.js";
+import { detectionNumClasses as a, resolveInputSize as o } from "../core/graph.js";
+import { modelNames as s } from "../core/metadata.js";
+import { OrtSession as c } from "../core/session.js";
+import { loadImage as l } from "../io/image.js";
+import { letterbox as u, toCHW as d, toFloat32 as f, toFloat32Tensor as p } from "../preprocess/image.js";
+import { decodeYolo as m } from "../postprocess/detection.js";
+import { VisionTask as h } from "./base.js";
 //#region src/vision/tasks/detector.ts
-var p = class p extends f {
+var g = class g extends h {
 	_head;
 	_labels;
 	_names;
@@ -22,9 +24,13 @@ var p = class p extends f {
 	static async create(e, t = {}) {
 		let n = t.head ?? "yolo";
 		if (n !== "yolo") throw Error(`Unsupported detector head '${n}'. Supported: 'yolo'.`);
-		let r = await a.create(e, t), o = i(t.labels ?? "coco", { numClasses: t.numClasses }), s = {};
-		for (let e = 0; e < o.length; e++) s[e] = o[e];
-		return new p(r, n, o, s, t.inputSize ?? [640, 640], t.confThreshold ?? .25, t.iouThreshold ?? .45, t.maxDetections ?? 300);
+		let r = await c.create(e, t), l = i(t.labels ?? s(r.metadata) ?? "coco", { numClasses: t.numClasses ?? a(r.outputShape) ?? void 0 }), u = {};
+		for (let e = 0; e < l.length; e++) u[e] = l[e];
+		return new g(r, n, l, u, o({
+			graphShape: r.inputShape,
+			requested: t.inputSize,
+			fallback: [640, 640]
+		}), t.confThreshold ?? .25, t.iouThreshold ?? .45, t.maxDetections ?? 300);
 	}
 	get head() {
 		return this._head;
@@ -35,6 +41,9 @@ var p = class p extends f {
 	get names() {
 		return this._names;
 	}
+	get inputSize() {
+		return this._inputSize;
+	}
 	get numClasses() {
 		return this._labels.length;
 	}
@@ -42,35 +51,35 @@ var p = class p extends f {
 		return this.predict(e, t);
 	}
 	async predict(e, n = {}) {
-		let i = new t(), a = typeof e == "string" ? e : null, s = await o(e);
+		let i = new t(), a = typeof e == "string" ? e : null, o = await l(e);
 		i.stage("load");
-		let { tensor: c, scale: l, padLeft: u, padTop: f } = this._preprocess(s);
+		let { tensor: s, scale: c, padLeft: u, padTop: d } = this._preprocess(o);
 		i.stage("preprocess");
-		let p = await this._session.run({ [this._session.inputName]: c });
+		let f = await this._session.run({ [this._session.inputName]: s });
 		i.stage("inference");
-		let m = this._session.outputNames[0];
-		if (m === void 0) throw Error("Detector model has no outputs.");
-		let h = p[m];
-		if (h === void 0) throw Error(`Detector model output ${m} missing from run() result.`);
-		let g = d(h.data, h.dims, {
-			originalWidth: s.width,
-			originalHeight: s.height,
+		let p = this._session.outputNames[0];
+		if (p === void 0) throw Error("Detector model has no outputs.");
+		let h = f[p];
+		if (h === void 0) throw Error(`Detector model output ${p} missing from run() result.`);
+		let g = m(h.data, h.dims, {
+			originalWidth: o.width,
+			originalHeight: o.height,
 			padLeft: u,
-			padTop: f,
-			scale: l,
+			padTop: d,
+			scale: c,
 			confThreshold: n.confThreshold ?? this._confThreshold,
 			iouThreshold: n.iouThreshold ?? this._iouThreshold,
 			maxDetections: this._maxDetections
 		}), _ = (n.classes === void 0 ? g : (() => {
 			let e = new Set(n.classes);
 			return g.filter((t) => e.has(t.classId));
-		})()).map((e) => this._buildResult(s, e.bbox, e.classId, e.confidence)), v = [s.height, s.width], y = this._buildBoxes(_, v);
-		return i.stage("postprocess"), [new r(y, _, this._names, s, v, a, i.speed())];
+		})()).map((e) => this._buildResult(o, e.bbox, e.classId, e.confidence)), v = [o.height, o.width], y = this._buildBoxes(_, v);
+		return i.stage("postprocess"), [new r(y, _, this._names, o, v, a, i.speed())];
 	}
 	_preprocess(e) {
-		let [t, n] = this._inputSize, r = s(e, t, n);
+		let [t, n] = this._inputSize, r = u(e, t, n);
 		return {
-			tensor: u(c(l(r.image), r.image.width, r.image.height, 3), [
+			tensor: p(d(f(r.image), r.image.width, r.image.height, 3), [
 				1,
 				3,
 				r.image.height,
@@ -114,6 +123,6 @@ var p = class p extends f {
 	}
 };
 //#endregion
-export { p as Detector };
+export { g as Detector };
 
 //# sourceMappingURL=detector.js.map

package/dist/vision/tasks/detector.js.map

@@ -1 +1 @@
-{"version":3,"file":"detector.js","names":[],"sources":["../../../src/vision/tasks/detector.ts"],"sourcesContent":["/**\n * Object detection task using anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYolo } from \"../postprocess/detection\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, DetectionResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type DetectionResult, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the model's detection head.\n *\n * - `\"yolo\"`: anchor-free YOLO head with output shape `[1, 4 + nc, N]` —\n *   covers YOLOv8, v9, v10, v11, v12, v26 detect exports.\n *\n * The SDK does **not** auto-detect the head from the model — the caller is\n * responsible for picking a head that matches their export. Future families\n * (v5/v6/v7 with `[1, N, 5+nc]`) will be added as new literal members.\n */\nexport type DetectorHead = \"yolo\";\n\nexport interface DetectorOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the detection head. Default `\"yolo\"` covers\n     * YOLOv8/v9/v10/v11/v12/v26.\n     */\n    readonly head?: DetectorHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /** Model input `[width, height]` for letterboxing. Defaults to `[640, 640]`. */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of detections per image. */\n    readonly maxDetections?: number;\n}\n\nexport interface DetectorPredictOptions {\n    /** Override the default confidence threshold. */\n    readonly confThreshold?: number;\n    /** Override the default IoU threshold. */\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only detections whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Object detector for anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n *\n * `predict()` returns `Promise<DetectionResults[]>` (length 1 for a single\n * image), mirroring Ultralytics' `YOLO(\"img.jpg\")`. Iterate the envelope for\n * per-instance dataclasses, or use the bulk `boxes` view (`.xyxy`, `.xywh`,\n * `.xyxyn`, `.xywhn`, `.cls`, `.conf`).\n *\n * @example\n * ```typescript\n * const det = await Detector.create(\"/models/yolov8n.onnx\");\n * const results = await det.predict(\"/images/street.jpg\");\n * const r = results[0];\n * console.log(r.boxes.xyxy, r.boxes.cls, r.boxes.conf, r.names);\n * for (const d of r) {\n *   console.log(d.cls, d.conf, d.box.xyxy);\n * }\n * ```\n */\nexport class Detector extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: DetectorHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: DetectorOptions = {}): Promise<Detector> {\n        const head: DetectorHead = options.head ?? \"yolo\";\n        if (head !== \"yolo\") {\n            throw new Error(`Unsupported detector head '${head}'. Supported: 'yolo'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? \"coco\", {\n            numClasses: options.numClasses,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Detector(\n            session,\n            head,\n            labels,\n            names,\n            options.inputSize ?? [640, 640],\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): DetectorHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /**\n     * Alias for {@link predict} — call the detector like a torch `nn.Module`.\n     *\n     * Use as `det.call(img)` since JavaScript class instances are not callable;\n     * for direct invocation, prefer `det.predict(img)`. The full\n     * {@link DetectorPredictOptions} (including `classes`) is supported.\n     */\n    async call(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        return this.predict(image, options);\n    }\n\n    /**\n     * Run detection on a single image.\n     *\n     * The returned envelope carries a {@link Speed} breakdown in `speed`,\n     * mirroring Ultralytics' `results[0].speed`.\n     */\n    async predict(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Detector model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(`Detector model output ${firstOutputName} missing from run() result.`);\n        }\n\n        const decodedAll = decodeYolo(raw.data as Float32Array, raw.dims, {\n            originalWidth: original.width,\n            originalHeight: original.height,\n            padLeft,\n            padTop,\n            scale,\n            confThreshold: options.confThreshold ?? this._confThreshold,\n            iouThreshold: options.iouThreshold ?? this._iouThreshold,\n            maxDetections: this._maxDetections,\n        });\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new DetectionResults(\n                boxes,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n    ): DetectionResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let cropped: RGBImage;\n        if (cx2 > cx1 && cy2 > cy1) {\n            const cw = cx2 - cx1;\n            const ch = cy2 - cy1;\n            const out = new Uint8Array(cw * ch * 3);\n            for (let row = 0; row < ch; row++) {\n                const srcOffset = ((cy1 + row) * original.width + cx1) * 3;\n                out.set(original.data.subarray(srcOffset, srcOffset + cw * 3), row * cw * 3);\n            }\n            cropped = new RGBImage(out, cw, ch);\n        } else {\n            cropped = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            croppedImage: cropped,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly DetectionResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as DetectionResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n}\n"],"mappings":";;;;;;;;;;AA+EA,IAAa,IAAb,MAAa,UAAiB,EAAW;CAGhB;CACA;CACA;CACA;CACA;CACA;CACA;CARrB,YACI,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACF;EADmB,AAEjB,MAAM,CAAO,GARI,KAAA,QAAA,GACA,KAAA,UAAA,GACA,KAAA,SAAA,GACA,KAAA,aAAA,GACA,KAAA,iBAAA,GACA,KAAA,gBAAA,GACA,KAAA,iBAAA;CAGrB;CAGA,aAAa,OAAO,GAAoB,IAA2B,CAAC,GAAsB;EACtF,IAAM,IAAqB,EAAQ,QAAQ;EAC3C,IAAI,MAAS,QACT,MAAU,MAAM,8BAA8B,EAAK,sBAAsB;EAE7E,IAAM,IAAU,MAAM,EAAW,OAAO,GAAO,CAAO,GAChD,IAAS,EAAc,EAAQ,UAAU,QAAQ,EACnD,YAAY,EAAQ,WACxB,CAAC,GACK,IAAgC,CAAC;EACvC,KAAK,IAAI,IAAI,GAAG,IAAI,EAAO,QAAQ,KAC/B,EAAM,KAAK,EAAO;EAEtB,OAAO,IAAI,EACP,GACA,GACA,GACA,GACA,EAAQ,aAAa,CAAC,KAAK,GAAG,GAC9B,EAAQ,iBAAiB,KACzB,EAAQ,gBAAgB,KACxB,EAAQ,iBAAiB,GAC7B;CACJ;CAGA,IAAI,OAAqB;EACrB,OAAO,KAAK;CAChB;CAGA,IAAI,SAA4B;EAC5B,OAAO,KAAK;CAChB;CAGA,IAAI,QAA0C;EAC1C,OAAO,KAAK;CAChB;CAGA,IAAI,aAAqB;EACrB,OAAO,KAAK,QAAQ;CACxB;CASA,MAAM,KACF,GACA,IAAkC,CAAC,GACR;EAC3B,OAAO,KAAK,QAAQ,GAAO,CAAO;CACtC;CAQA,MAAM,QACF,GACA,IAAkC,CAAC,GACR;EAC3B,IAAM,IAAQ,IAAI,EAAW,GACvB,IAAO,OAAO,KAAU,WAAW,IAAQ,MAC3C,IAAW,MAAM,EAAU,CAAK;EACtC,EAAM,MAAM,MAAM;EAClB,IAAM,EAAE,WAAQ,UAAO,YAAS,cAAW,KAAK,YAAY,CAAQ;EACpE,EAAM,MAAM,YAAY;EACxB,IAAM,IAAU,MAAM,KAAK,SAAS,IAAI,GAAG,KAAK,SAAS,YAAY,EAAO,CAAC;EAC7E,EAAM,MAAM,WAAW;EAEvB,IAAM,IAAkB,KAAK,SAAS,YAAY;EAClD,IAAI,MAAoB,KAAA,GACpB,MAAU,MAAM,gCAAgC;EAEpD,IAAM,IAAM,EAAQ;EACpB,IAAI,MAAQ,KAAA,GACR,MAAU,MAAM,yBAAyB,EAAgB,4BAA4B;EAGzF,IAAM,IAAa,EAAW,EAAI,MAAsB,EAAI,MAAM;GAC9D,eAAe,EAAS;GACxB,gBAAgB,EAAS;GACzB;GACA;GACA;GACA,eAAe,EAAQ,iBAAiB,KAAK;GAC7C,cAAc,EAAQ,gBAAgB,KAAK;GAC3C,eAAe,KAAK;EACxB,CAAC,GAUK,KAPF,EAAQ,YAAY,KAAA,IAKd,WAJO;GACH,IAAM,IAAU,IAAI,IAAI,EAAQ,OAAO;GACvC,OAAO,EAAW,QAAQ,MAAM,EAAQ,IAAI,EAAE,OAAO,CAAC;EAC1D,EAAA,CAAG,EACH,CAEiB,KAAK,MAC5B,KAAK,aAAa,GAAU,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,CAC/D,GAEM,IAAkC,CAAC,EAAS,QAAQ,EAAS,KAAK,GAClE,IAAQ,KAAK,YAAY,GAAY,CAAI;EAE/C,OADA,EAAM,MAAM,aAAa,GAClB,CACH,IAAI,EACA,GACA,GACA,KAAK,QACL,GACA,GACA,GACA,EAAM,MAAM,CAChB,CACJ;CACJ;CAEA,YAAoB,GAKlB;EACE,IAAM,CAAC,GAAI,KAAM,KAAK,YAChB,IAAK,EAAU,GAAO,GAAI,CAAE;EAGlC,OAAO;GACH,QAAQ,EAFA,EADA,EAAU,EAAG,KACP,GAAK,EAAG,MAAM,OAAO,EAAG,MAAM,QAAQ,CAE5B,GAAK;IAAC;IAAG;IAAG,EAAG,MAAM;IAAQ,EAAG,MAAM;GAAK,CAAC;GACpE,OAAO,EAAG;GACV,SAAS,EAAG;GACZ,QAAQ,EAAG;EACf;CACJ;CAEA,aACI,GACA,GACA,GACA,GACe;EACf,IAAM,CAAC,GAAI,GAAI,GAAI,KAAM,EAAK,UAAU,GAClC,IAAM,KAAK,IAAI,GAAG,CAAE,GACpB,IAAM,KAAK,IAAI,GAAG,CAAE,GACpB,IAAM,KAAK,IAAI,EAAS,OAAO,CAAE,GACjC,IAAM,KAAK,IAAI,EAAS,QAAQ,CAAE,GAEpC;EACJ,IAAI,IAAM,KAAO,IAAM,GAAK;GACxB,IAAM,IAAK,IAAM,GACX,IAAK,IAAM,GACX,IAAM,IAAI,WAAW,IAAK,IAAK,CAAC;GACtC,KAAK,IAAI,IAAM,GAAG,IAAM,GAAI,KAAO;IAC/B,IAAM,MAAc,IAAM,KAAO,EAAS,QAAQ,KAAO;IACzD,EAAI,IAAI,EAAS,KAAK,SAAS,GAAW,IAAY,IAAK,CAAC,GAAG,IAAM,IAAK,CAAC;GAC/E;GACA,IAAU,IAAI,EAAS,GAAK,GAAI,CAAE;EACtC,OACI,IAAU,IAAI,kBAAS,IAAI,WAAY,GAAG,GAAG,CAAC;EAGlD,IAAM,IAAY,KAAK,OAAO,MAAY,SAAS;EAEnD,OAAO;GACH;GACA;GACA;GACA;GACA,KAAK;GACL,MAAM;GACN,MAAM;GACN,KAAK;GACL,cAAc;EAClB;CACJ;CAEA,YACI,GACA,GACK;EACL,IAAM,IAAI,EAAW,QACf,IAAO,IAAI,aAAa,IAAI,CAAC,GAC7B,IAAM,IAAI,WAAW,CAAC,GACtB,IAAO,IAAI,aAAa,CAAC;EAC/B,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KAAK;GACxB,IAAM,IAAI,EAAW;GAMrB,AALA,EAAK,IAAI,KAAK,EAAE,KAAK,IACrB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAI,KAAK,EAAE,SACX,EAAK,KAAK,EAAE;EAChB;EACA,OAAO,IAAI,EAAM,GAAM,GAAK,GAAM,CAAS;CAC/C;AACJ"}
+{"version":3,"file":"detector.js","names":[],"sources":["../../../src/vision/tasks/detector.ts"],"sourcesContent":["/**\n * Object detection task using anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\n\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { detectionNumClasses, resolveInputSize } from \"../core/graph\";\nimport { modelNames } from \"../core/metadata\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYolo } from \"../postprocess/detection\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, DetectionResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type DetectionResult, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the model's detection head.\n *\n * - `\"yolo\"`: anchor-free YOLO head with output shape `[1, 4 + nc, N]` —\n *   covers YOLOv8, v9, v10, v11, v12, v26 detect exports.\n *\n * The SDK does **not** auto-detect the head from the model — the caller is\n * responsible for picking a head that matches their export. Future families\n * (v5/v6/v7 with `[1, N, 5+nc]`) will be added as new literal members.\n */\nexport type DetectorHead = \"yolo\";\n\nexport interface DetectorOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the detection head. Default `\"yolo\"` covers\n     * YOLOv8/v9/v10/v11/v12/v26.\n     */\n    readonly head?: DetectorHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /**\n     * Model input `[width, height]` in pixels for letterboxing.\n     *\n     * Only used when the model's graph leaves its spatial axes dynamic: a graph\n     * that declares a static size always wins, since that is the only shape ONNX\n     * Runtime will accept. Defaults to `[640, 640]`.\n     */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of detections per image. */\n    readonly maxDetections?: number;\n}\n\nexport interface DetectorPredictOptions {\n    /** Override the default confidence threshold. */\n    readonly confThreshold?: number;\n    /** Override the default IoU threshold. */\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only detections whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Object detector for anchor-free YOLO ONNX models (v8/v9/v10/v11/v12).\n *\n * `predict()` returns `Promise<DetectionResults[]>` (length 1 for a single\n * image), mirroring Ultralytics' `YOLO(\"img.jpg\")`. Iterate the envelope for\n * per-instance dataclasses, or use the bulk `boxes` view (`.xyxy`, `.xywh`,\n * `.xyxyn`, `.xywhn`, `.cls`, `.conf`).\n *\n * @example\n * ```typescript\n * const det = await Detector.create(\"/models/yolov8n.onnx\");\n * const results = await det.predict(\"/images/street.jpg\");\n * const r = results[0];\n * console.log(r.boxes.xyxy, r.boxes.cls, r.boxes.conf, r.names);\n * for (const d of r) {\n *   console.log(d.cls, d.conf, d.box.xyxy);\n * }\n * ```\n */\nexport class Detector extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: DetectorHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: DetectorOptions = {}): Promise<Detector> {\n        const head: DetectorHead = options.head ?? \"yolo\";\n        if (head !== \"yolo\") {\n            throw new Error(`Unsupported detector head '${head}'. Supported: 'yolo'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? modelNames(session.metadata) ?? \"coco\", {\n            numClasses: options.numClasses ?? detectionNumClasses(session.outputShape) ?? undefined,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Detector(\n            session,\n            head,\n            labels,\n            names,\n            resolveInputSize({\n                graphShape: session.inputShape,\n                requested: options.inputSize,\n                fallback: [640, 640],\n            }),\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): DetectorHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /**\n     * The `[width, height]` this task preprocesses to.\n     *\n     * Resolved at creation time from the model's graph when it declares a static\n     * input, so reading it back tells you the resolution inference really runs at\n     * — not merely what was requested.\n     */\n    get inputSize(): readonly [number, number] {\n        return this._inputSize;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /**\n     * Alias for {@link predict} — call the detector like a torch `nn.Module`.\n     *\n     * Use as `det.call(img)` since JavaScript class instances are not callable;\n     * for direct invocation, prefer `det.predict(img)`. The full\n     * {@link DetectorPredictOptions} (including `classes`) is supported.\n     */\n    async call(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        return this.predict(image, options);\n    }\n\n    /**\n     * Run detection on a single image.\n     *\n     * The returned envelope carries a {@link Speed} breakdown in `speed`,\n     * mirroring Ultralytics' `results[0].speed`.\n     */\n    async predict(\n        image: ImageInput,\n        options: DetectorPredictOptions = {},\n    ): Promise<DetectionResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const firstOutputName = this._session.outputNames[0];\n        if (firstOutputName === undefined) {\n            throw new Error(\"Detector model has no outputs.\");\n        }\n        const raw = outputs[firstOutputName];\n        if (raw === undefined) {\n            throw new Error(`Detector model output ${firstOutputName} missing from run() result.`);\n        }\n\n        const decodedAll = decodeYolo(raw.data as Float32Array, raw.dims, {\n            originalWidth: original.width,\n            originalHeight: original.height,\n            padLeft,\n            padTop,\n            scale,\n            confThreshold: options.confThreshold ?? this._confThreshold,\n            iouThreshold: options.iouThreshold ?? this._iouThreshold,\n            maxDetections: this._maxDetections,\n        });\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new DetectionResults(\n                boxes,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n    ): DetectionResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let cropped: RGBImage;\n        if (cx2 > cx1 && cy2 > cy1) {\n            const cw = cx2 - cx1;\n            const ch = cy2 - cy1;\n            const out = new Uint8Array(cw * ch * 3);\n            for (let row = 0; row < ch; row++) {\n                const srcOffset = ((cy1 + row) * original.width + cx1) * 3;\n                out.set(original.data.subarray(srcOffset, srcOffset + cw * 3), row * cw * 3);\n            }\n            cropped = new RGBImage(out, cw, ch);\n        } else {\n            cropped = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            croppedImage: cropped,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly DetectionResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as DetectionResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n}\n"],"mappings":";;;;;;;;;;;;AAwFA,IAAa,IAAb,MAAa,UAAiB,EAAW;CAGhB;CACA;CACA;CACA;CACA;CACA;CACA;CARrB,YACI,GACA,GACA,GACA,GACA,GACA,GACA,GACA,GACF;EADmB,AAEjB,MAAM,CAAO,GARI,KAAA,QAAA,GACA,KAAA,UAAA,GACA,KAAA,SAAA,GACA,KAAA,aAAA,GACA,KAAA,iBAAA,GACA,KAAA,gBAAA,GACA,KAAA,iBAAA;CAGrB;CAGA,aAAa,OAAO,GAAoB,IAA2B,CAAC,GAAsB;EACtF,IAAM,IAAqB,EAAQ,QAAQ;EAC3C,IAAI,MAAS,QACT,MAAU,MAAM,8BAA8B,EAAK,sBAAsB;EAE7E,IAAM,IAAU,MAAM,EAAW,OAAO,GAAO,CAAO,GAChD,IAAS,EAAc,EAAQ,UAAU,EAAW,EAAQ,QAAQ,KAAK,QAAQ,EACnF,YAAY,EAAQ,cAAc,EAAoB,EAAQ,WAAW,KAAK,KAAA,EAClF,CAAC,GACK,IAAgC,CAAC;EACvC,KAAK,IAAI,IAAI,GAAG,IAAI,EAAO,QAAQ,KAC/B,EAAM,KAAK,EAAO;EAEtB,OAAO,IAAI,EACP,GACA,GACA,GACA,GACA,EAAiB;GACb,YAAY,EAAQ;GACpB,WAAW,EAAQ;GACnB,UAAU,CAAC,KAAK,GAAG;EACvB,CAAC,GACD,EAAQ,iBAAiB,KACzB,EAAQ,gBAAgB,KACxB,EAAQ,iBAAiB,GAC7B;CACJ;CAGA,IAAI,OAAqB;EACrB,OAAO,KAAK;CAChB;CAGA,IAAI,SAA4B;EAC5B,OAAO,KAAK;CAChB;CAGA,IAAI,QAA0C;EAC1C,OAAO,KAAK;CAChB;CASA,IAAI,YAAuC;EACvC,OAAO,KAAK;CAChB;CAGA,IAAI,aAAqB;EACrB,OAAO,KAAK,QAAQ;CACxB;CASA,MAAM,KACF,GACA,IAAkC,CAAC,GACR;EAC3B,OAAO,KAAK,QAAQ,GAAO,CAAO;CACtC;CAQA,MAAM,QACF,GACA,IAAkC,CAAC,GACR;EAC3B,IAAM,IAAQ,IAAI,EAAW,GACvB,IAAO,OAAO,KAAU,WAAW,IAAQ,MAC3C,IAAW,MAAM,EAAU,CAAK;EACtC,EAAM,MAAM,MAAM;EAClB,IAAM,EAAE,WAAQ,UAAO,YAAS,cAAW,KAAK,YAAY,CAAQ;EACpE,EAAM,MAAM,YAAY;EACxB,IAAM,IAAU,MAAM,KAAK,SAAS,IAAI,GAAG,KAAK,SAAS,YAAY,EAAO,CAAC;EAC7E,EAAM,MAAM,WAAW;EAEvB,IAAM,IAAkB,KAAK,SAAS,YAAY;EAClD,IAAI,MAAoB,KAAA,GACpB,MAAU,MAAM,gCAAgC;EAEpD,IAAM,IAAM,EAAQ;EACpB,IAAI,MAAQ,KAAA,GACR,MAAU,MAAM,yBAAyB,EAAgB,4BAA4B;EAGzF,IAAM,IAAa,EAAW,EAAI,MAAsB,EAAI,MAAM;GAC9D,eAAe,EAAS;GACxB,gBAAgB,EAAS;GACzB;GACA;GACA;GACA,eAAe,EAAQ,iBAAiB,KAAK;GAC7C,cAAc,EAAQ,gBAAgB,KAAK;GAC3C,eAAe,KAAK;EACxB,CAAC,GAUK,KAPF,EAAQ,YAAY,KAAA,IAKd,WAJO;GACH,IAAM,IAAU,IAAI,IAAI,EAAQ,OAAO;GACvC,OAAO,EAAW,QAAQ,MAAM,EAAQ,IAAI,EAAE,OAAO,CAAC;EAC1D,EAAA,CAAG,EACH,CAEiB,KAAK,MAC5B,KAAK,aAAa,GAAU,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,CAC/D,GAEM,IAAkC,CAAC,EAAS,QAAQ,EAAS,KAAK,GAClE,IAAQ,KAAK,YAAY,GAAY,CAAI;EAE/C,OADA,EAAM,MAAM,aAAa,GAClB,CACH,IAAI,EACA,GACA,GACA,KAAK,QACL,GACA,GACA,GACA,EAAM,MAAM,CAChB,CACJ;CACJ;CAEA,YAAoB,GAKlB;EACE,IAAM,CAAC,GAAI,KAAM,KAAK,YAChB,IAAK,EAAU,GAAO,GAAI,CAAE;EAGlC,OAAO;GACH,QAAQ,EAFA,EADA,EAAU,EAAG,KACP,GAAK,EAAG,MAAM,OAAO,EAAG,MAAM,QAAQ,CAE5B,GAAK;IAAC;IAAG;IAAG,EAAG,MAAM;IAAQ,EAAG,MAAM;GAAK,CAAC;GACpE,OAAO,EAAG;GACV,SAAS,EAAG;GACZ,QAAQ,EAAG;EACf;CACJ;CAEA,aACI,GACA,GACA,GACA,GACe;EACf,IAAM,CAAC,GAAI,GAAI,GAAI,KAAM,EAAK,UAAU,GAClC,IAAM,KAAK,IAAI,GAAG,CAAE,GACpB,IAAM,KAAK,IAAI,GAAG,CAAE,GACpB,IAAM,KAAK,IAAI,EAAS,OAAO,CAAE,GACjC,IAAM,KAAK,IAAI,EAAS,QAAQ,CAAE,GAEpC;EACJ,IAAI,IAAM,KAAO,IAAM,GAAK;GACxB,IAAM,IAAK,IAAM,GACX,IAAK,IAAM,GACX,IAAM,IAAI,WAAW,IAAK,IAAK,CAAC;GACtC,KAAK,IAAI,IAAM,GAAG,IAAM,GAAI,KAAO;IAC/B,IAAM,MAAc,IAAM,KAAO,EAAS,QAAQ,KAAO;IACzD,EAAI,IAAI,EAAS,KAAK,SAAS,GAAW,IAAY,IAAK,CAAC,GAAG,IAAM,IAAK,CAAC;GAC/E;GACA,IAAU,IAAI,EAAS,GAAK,GAAI,CAAE;EACtC,OACI,IAAU,IAAI,kBAAS,IAAI,WAAY,GAAG,GAAG,CAAC;EAGlD,IAAM,IAAY,KAAK,OAAO,MAAY,SAAS;EAEnD,OAAO;GACH;GACA;GACA;GACA;GACA,KAAK;GACL,MAAM;GACN,MAAM;GACN,KAAK;GACL,cAAc;EAClB;CACJ;CAEA,YACI,GACA,GACK;EACL,IAAM,IAAI,EAAW,QACf,IAAO,IAAI,aAAa,IAAI,CAAC,GAC7B,IAAM,IAAI,WAAW,CAAC,GACtB,IAAO,IAAI,aAAa,CAAC;EAC/B,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KAAK;GACxB,IAAM,IAAI,EAAW;GAMrB,AALA,EAAK,IAAI,KAAK,EAAE,KAAK,IACrB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAK,IAAI,IAAI,KAAK,EAAE,KAAK,IACzB,EAAI,KAAK,EAAE,SACX,EAAK,KAAK,EAAE;EAChB;EACA,OAAO,IAAI,EAAM,GAAM,GAAK,GAAM,CAAS;CAC/C;AACJ"}

package/dist/vision/tasks/segmenter.cjs

@@ -1,2 +1,2 @@
-const e=require("../types.cjs"),t=require("../core/timing.cjs"),n=require("../results.cjs"),r=require("../labels.cjs"),i=require("../core/session.cjs"),a=require("../io/image.cjs"),o=require("../preprocess/image.cjs"),s=require("../postprocess/segmentation.cjs"),c=require("./base.cjs");var l=class l extends c.VisionTask{_head;_labels;_names;_inputSize;_confThreshold;_iouThreshold;_maxDetections;_maskThreshold;constructor(e,t,n,r,i,a,o,s,c){super(e),this._head=t,this._labels=n,this._names=r,this._inputSize=i,this._confThreshold=a,this._iouThreshold=o,this._maxDetections=s,this._maskThreshold=c}static async create(e,t={}){let n=t.head??`yolo-seg`;if(n!==`yolo-seg`)throw Error(`Unsupported segmenter head '${n}'. Supported: 'yolo-seg'.`);let a=await i.OrtSession.create(e,t),o=r.resolveLabels(t.labels??`coco`,{numClasses:t.numClasses}),s={};for(let e=0;e<o.length;e++)s[e]=o[e];return new l(a,n,o,s,t.inputSize??[640,640],t.confThreshold??.25,t.iouThreshold??.45,t.maxDetections??300,t.maskThreshold??.5)}get head(){return this._head}get labels(){return this._labels}get names(){return this._names}get numClasses(){return this._labels.length}async call(e,t={}){return this.predict(e,t)}async predict(e,r={}){let i=new t.SpeedTimer,o=typeof e==`string`?e:null,c=await a.loadImage(e);i.stage(`load`);let{tensor:l,scale:u,padLeft:d,padTop:f}=this._preprocess(c);i.stage(`preprocess`);let p=await this._session.run({[this._session.inputName]:l});i.stage(`inference`);let{perAnchor:m,prototypes:h}=this._splitOutputs(p),g=s.decodeYoloSeg(m.data,m.dims,h.data,h.dims,{numClasses:this._labels.length,inputWidth:this._inputSize[0],inputHeight:this._inputSize[1],originalWidth:c.width,originalHeight:c.height,padLeft:d,padTop:f,scale:u,confThreshold:r.confThreshold??this._confThreshold,iouThreshold:r.iouThreshold??this._iouThreshold,maxDetections:this._maxDetections,maskThreshold:this._maskThreshold}),_=(r.classes===void 0?g:(()=>{let e=new Set(r.classes);return g.filter(t=>e.has(t.classId))})()).map(e=>this._buildResult(c,e.bbox,e.classId,e.confidence,e.mask)),v=[c.height,c.width],y=this._buildBoxes(_,v),b=this._buildMasks(_,v);return i.stage(`postprocess`),[new n.SegmentationResults(y,b,_,this._names,c,v,o,i.speed())]}_preprocess(e){let[t,n]=this._inputSize,r=o.letterbox(e,t,n);return{tensor:o.toFloat32Tensor(o.toCHW(o.toFloat32(r.image),r.image.width,r.image.height,3),[1,3,r.image.height,r.image.width]),scale:r.scale,padLeft:r.padLeft,padTop:r.padTop}}_splitOutputs(e){let t,n;for(let r of this._session.outputNames){let i=e[r];i!==void 0&&(i.dims.length===3&&t===void 0?t=i:i.dims.length===4&&n===void 0&&(n=i))}if(t===void 0||n===void 0){let t=this._session.outputNames.map(t=>`${t}: ${JSON.stringify(e[t]?.dims??[])}`);throw Error(`Segmenter expected one 3-D and one 4-D output, got [${t.join(`, `)}].`)}return{perAnchor:t,prototypes:n}}_buildResult(t,n,r,i,a){let[o,s,c,l]=n.asIntXyxy(),u=Math.max(0,o),d=Math.max(0,s),f=Math.min(t.width,c),p=Math.min(t.height,l),m,h=a;if(f>u&&p>d&&a.data.length>0){let n=f-u,r=p-d,i=Math.min(a.width,n),o=Math.min(a.height,r),s=new Uint8Array(i*o*3);for(let e=0;e<o;e++){let n=((d+e)*t.width+u)*3,r=e*i*3,o=e*a.width;for(let e=0;e<i;e++)if(a.data[o+e]!==0){let i=n+e*3,a=r+e*3;s[a]=t.data[i],s[a+1]=t.data[i+1],s[a+2]=t.data[i+2]}}if(m=new e.RGBImage(s,i,o),i!==a.width||o!==a.height){let t=new Uint8Array(i*o);for(let e=0;e<o;e++)t.set(a.data.subarray(e*a.width,e*a.width+i),e*i);h=new e.Mask(t,i,o)}}else h=new e.Mask(new Uint8Array,0,0),m=new e.RGBImage(new Uint8Array,0,0);let g=this._names[r]??`class_${r}`;return{classId:r,className:g,confidence:i,bbox:n,cls:r,name:g,conf:i,box:n,mask:h,segmentedImage:m}}_buildBoxes(e,t){let r=e.length,i=new Float32Array(r*4),a=new Int32Array(r),o=new Float32Array(r);for(let t=0;t<r;t++){let n=e[t];i[t*4]=n.bbox.x1,i[t*4+1]=n.bbox.y1,i[t*4+2]=n.bbox.x2,i[t*4+3]=n.bbox.y2,a[t]=n.classId,o[t]=n.confidence}return new n.Boxes(i,a,o,t)}_buildMasks(e,t){let r=new Float32Array(e.length*4);for(let t=0;t<e.length;t++){let n=e[t];r[t*4]=n.bbox.x1,r[t*4+1]=n.bbox.y1,r[t*4+2]=n.bbox.x2,r[t*4+3]=n.bbox.y2}return new n.Masks(e.map(e=>e.mask),r,t)}};exports.Segmenter=l;
+const e=require("../types.cjs"),t=require("../core/timing.cjs"),n=require("../results.cjs"),r=require("../labels.cjs"),i=require("../core/graph.cjs"),a=require("../core/metadata.cjs"),o=require("../core/session.cjs"),s=require("../io/image.cjs"),c=require("../preprocess/image.cjs"),l=require("../postprocess/segmentation.cjs"),u=require("./base.cjs");var d=class d extends u.VisionTask{_head;_labels;_names;_inputSize;_confThreshold;_iouThreshold;_maxDetections;_maskThreshold;constructor(e,t,n,r,i,a,o,s,c){super(e),this._head=t,this._labels=n,this._names=r,this._inputSize=i,this._confThreshold=a,this._iouThreshold=o,this._maxDetections=s,this._maskThreshold=c}static async create(e,t={}){let n=t.head??`yolo-seg`;if(n!==`yolo-seg`)throw Error(`Unsupported segmenter head '${n}'. Supported: 'yolo-seg'.`);let s=await o.OrtSession.create(e,t),c=r.resolveLabels(t.labels??a.modelNames(s.metadata)??`coco`,{numClasses:t.numClasses??i.detectionNumClasses(s.outputShape)??void 0}),l={};for(let e=0;e<c.length;e++)l[e]=c[e];return new d(s,n,c,l,i.resolveInputSize({graphShape:s.inputShape,requested:t.inputSize,fallback:[640,640]}),t.confThreshold??.25,t.iouThreshold??.45,t.maxDetections??300,t.maskThreshold??.5)}get head(){return this._head}get labels(){return this._labels}get names(){return this._names}get inputSize(){return this._inputSize}get numClasses(){return this._labels.length}async call(e,t={}){return this.predict(e,t)}async predict(e,r={}){let i=new t.SpeedTimer,a=typeof e==`string`?e:null,o=await s.loadImage(e);i.stage(`load`);let{tensor:c,scale:u,padLeft:d,padTop:f}=this._preprocess(o);i.stage(`preprocess`);let p=await this._session.run({[this._session.inputName]:c});i.stage(`inference`);let{perAnchor:m,prototypes:h}=this._splitOutputs(p),g=l.decodeYoloSeg(m.data,m.dims,h.data,h.dims,{numClasses:this._labels.length,inputWidth:this._inputSize[0],inputHeight:this._inputSize[1],originalWidth:o.width,originalHeight:o.height,padLeft:d,padTop:f,scale:u,confThreshold:r.confThreshold??this._confThreshold,iouThreshold:r.iouThreshold??this._iouThreshold,maxDetections:this._maxDetections,maskThreshold:this._maskThreshold}),_=(r.classes===void 0?g:(()=>{let e=new Set(r.classes);return g.filter(t=>e.has(t.classId))})()).map(e=>this._buildResult(o,e.bbox,e.classId,e.confidence,e.mask)),v=[o.height,o.width],y=this._buildBoxes(_,v),b=this._buildMasks(_,v);return i.stage(`postprocess`),[new n.SegmentationResults(y,b,_,this._names,o,v,a,i.speed())]}_preprocess(e){let[t,n]=this._inputSize,r=c.letterbox(e,t,n);return{tensor:c.toFloat32Tensor(c.toCHW(c.toFloat32(r.image),r.image.width,r.image.height,3),[1,3,r.image.height,r.image.width]),scale:r.scale,padLeft:r.padLeft,padTop:r.padTop}}_splitOutputs(e){let t,n;for(let r of this._session.outputNames){let i=e[r];i!==void 0&&(i.dims.length===3&&t===void 0?t=i:i.dims.length===4&&n===void 0&&(n=i))}if(t===void 0||n===void 0){let t=this._session.outputNames.map(t=>`${t}: ${JSON.stringify(e[t]?.dims??[])}`);throw Error(`Segmenter expected one 3-D and one 4-D output, got [${t.join(`, `)}].`)}return{perAnchor:t,prototypes:n}}_buildResult(t,n,r,i,a){let[o,s,c,l]=n.asIntXyxy(),u=Math.max(0,o),d=Math.max(0,s),f=Math.min(t.width,c),p=Math.min(t.height,l),m,h=a;if(f>u&&p>d&&a.data.length>0){let n=f-u,r=p-d,i=Math.min(a.width,n),o=Math.min(a.height,r),s=new Uint8Array(i*o*3);for(let e=0;e<o;e++){let n=((d+e)*t.width+u)*3,r=e*i*3,o=e*a.width;for(let e=0;e<i;e++)if(a.data[o+e]!==0){let i=n+e*3,a=r+e*3;s[a]=t.data[i],s[a+1]=t.data[i+1],s[a+2]=t.data[i+2]}}if(m=new e.RGBImage(s,i,o),i!==a.width||o!==a.height){let t=new Uint8Array(i*o);for(let e=0;e<o;e++)t.set(a.data.subarray(e*a.width,e*a.width+i),e*i);h=new e.Mask(t,i,o)}}else h=new e.Mask(new Uint8Array,0,0),m=new e.RGBImage(new Uint8Array,0,0);let g=this._names[r]??`class_${r}`;return{classId:r,className:g,confidence:i,bbox:n,cls:r,name:g,conf:i,box:n,mask:h,segmentedImage:m}}_buildBoxes(e,t){let r=e.length,i=new Float32Array(r*4),a=new Int32Array(r),o=new Float32Array(r);for(let t=0;t<r;t++){let n=e[t];i[t*4]=n.bbox.x1,i[t*4+1]=n.bbox.y1,i[t*4+2]=n.bbox.x2,i[t*4+3]=n.bbox.y2,a[t]=n.classId,o[t]=n.confidence}return new n.Boxes(i,a,o,t)}_buildMasks(e,t){let r=new Float32Array(e.length*4);for(let t=0;t<e.length;t++){let n=e[t];r[t*4]=n.bbox.x1,r[t*4+1]=n.bbox.y1,r[t*4+2]=n.bbox.x2,r[t*4+3]=n.bbox.y2}return new n.Masks(e.map(e=>e.mask),r,t)}};exports.Segmenter=d;
 //# sourceMappingURL=segmenter.cjs.map

package/dist/vision/tasks/segmenter.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"segmenter.cjs","names":[],"sources":["../../../src/vision/tasks/segmenter.ts"],"sourcesContent":["/**\n * Instance-segmentation task using YOLO seg ONNX models (v8-seg / v11-seg / ...).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYoloSeg } from \"../postprocess/segmentation\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, Masks, SegmentationResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type SegmentationResult, Mask, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the segmentation head.\n *\n * - `\"yolo-seg\"`: YOLO instance-segmentation head with two outputs —\n *   `[1, 4 + nc + nm, N]` per-anchor predictions plus `[1, nm, mh, mw]`\n *   prototype masks. Covers YOLOv8-seg, v11-seg, v26-seg.\n *\n * The SDK does **not** auto-detect this — the caller is responsible for\n * picking a head that matches their export.\n */\nexport type SegmenterHead = \"yolo-seg\";\n\nexport interface SegmenterOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the segmentation head. Default `\"yolo-seg\"` covers\n     * YOLOv8-seg/v11-seg/v26-seg.\n     */\n    readonly head?: SegmenterHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /** Model input `[width, height]` for letterboxing. Defaults to `[640, 640]`. */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of instances per image. */\n    readonly maxDetections?: number;\n    /** Probability cutoff applied to soft masks. Defaults to `0.5`. */\n    readonly maskThreshold?: number;\n}\n\nexport interface SegmenterPredictOptions {\n    readonly confThreshold?: number;\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only instances whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Instance segmenter for YOLO seg ONNX models (v8-seg / v11-seg / ...).\n *\n * The model is expected to expose two outputs:\n *\n * 1. `output0`: `(1, 4 + numClasses + numMaskCoefs, numAnchors)` — per-anchor\n *    predictions (boxes, class scores, mask coefficients).\n * 2. `output1`: `(1, numMaskCoefs, maskH, maskW)` — prototype masks.\n *\n * `predict()` returns `Promise<SegmentationResults[]>` (length 1 for a\n * single image), mirroring Ultralytics' API. The envelope exposes:\n *\n * - `boxes`: bulk numpy view (`xyxy`, `xywh`, `xyxyn`, `xywhn`, `cls`, `conf`).\n * - `masks`: per-instance binary masks cropped to each box.\n * - per-instance {@link SegmentationResult} via iteration.\n *\n * @example\n * ```typescript\n * const seg = await Segmenter.create(\"/models/yolov8n-seg.onnx\");\n * const r = (await seg.predict(\"/images/street.jpg\"))[0];\n * for (const inst of r) {\n *   console.log(inst.cls, inst.conf, inst.box.xyxy);\n * }\n * ```\n */\nexport class Segmenter extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: SegmenterHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n        private readonly _maskThreshold: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: SegmenterOptions = {}): Promise<Segmenter> {\n        const head: SegmenterHead = options.head ?? \"yolo-seg\";\n        if (head !== \"yolo-seg\") {\n            throw new Error(`Unsupported segmenter head '${head}'. Supported: 'yolo-seg'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? \"coco\", {\n            numClasses: options.numClasses,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Segmenter(\n            session,\n            head,\n            labels,\n            names,\n            options.inputSize ?? [640, 640],\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n            options.maskThreshold ?? 0.5,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): SegmenterHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /** Alias for {@link predict} (parity with PyTorch `nn.Module.__call__`). */\n    async call(\n        image: ImageInput,\n        options: SegmenterPredictOptions = {},\n    ): Promise<SegmentationResults[]> {\n        return this.predict(image, options);\n    }\n\n    /** Run instance segmentation on a single image. */\n    async predict(\n        image: ImageInput,\n        options: SegmenterPredictOptions = {},\n    ): Promise<SegmentationResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const { perAnchor, prototypes } = this._splitOutputs(outputs);\n\n        const decodedAll = decodeYoloSeg(\n            perAnchor.data as Float32Array,\n            perAnchor.dims,\n            prototypes.data as Float32Array,\n            prototypes.dims,\n            {\n                numClasses: this._labels.length,\n                inputWidth: this._inputSize[0],\n                inputHeight: this._inputSize[1],\n                originalWidth: original.width,\n                originalHeight: original.height,\n                padLeft,\n                padTop,\n                scale,\n                confThreshold: options.confThreshold ?? this._confThreshold,\n                iouThreshold: options.iouThreshold ?? this._iouThreshold,\n                maxDetections: this._maxDetections,\n                maskThreshold: this._maskThreshold,\n            },\n        );\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence, d.mask),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        const masks = this._buildMasks(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new SegmentationResults(\n                boxes,\n                masks,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _splitOutputs(outputs: Record<string, ort.Tensor>): {\n        perAnchor: ort.Tensor;\n        prototypes: ort.Tensor;\n    } {\n        let perAnchor: ort.Tensor | undefined;\n        let prototypes: ort.Tensor | undefined;\n        for (const name of this._session.outputNames) {\n            const t = outputs[name];\n            if (t === undefined) continue;\n            if (t.dims.length === 3 && perAnchor === undefined) {\n                perAnchor = t;\n            } else if (t.dims.length === 4 && prototypes === undefined) {\n                prototypes = t;\n            }\n        }\n        if (perAnchor === undefined || prototypes === undefined) {\n            const shapes = this._session.outputNames.map(\n                (n) => `${n}: ${JSON.stringify(outputs[n]?.dims ?? [])}`,\n            );\n            throw new Error(\n                `Segmenter expected one 3-D and one 4-D output, got [${shapes.join(\", \")}].`,\n            );\n        }\n        return { perAnchor, prototypes };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n        mask: Mask,\n    ): SegmentationResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let segmentedImage: RGBImage;\n        let finalMask = mask;\n        if (cx2 > cx1 && cy2 > cy1 && mask.data.length > 0) {\n            const cropW = cx2 - cx1;\n            const cropH = cy2 - cy1;\n            const mw = Math.min(mask.width, cropW);\n            const mh = Math.min(mask.height, cropH);\n            const segData = new Uint8Array(mw * mh * 3);\n            for (let row = 0; row < mh; row++) {\n                const srcRowOffset = ((cy1 + row) * original.width + cx1) * 3;\n                const dstRowOffset = row * mw * 3;\n                const maskRowOffset = row * mask.width;\n                for (let col = 0; col < mw; col++) {\n                    const m = mask.data[maskRowOffset + col];\n                    if (m !== 0) {\n                        const s = srcRowOffset + col * 3;\n                        const d = dstRowOffset + col * 3;\n                        segData[d] = original.data[s];\n                        segData[d + 1] = original.data[s + 1];\n                        segData[d + 2] = original.data[s + 2];\n                    }\n                }\n            }\n            segmentedImage = new RGBImage(segData, mw, mh);\n            if (mw !== mask.width || mh !== mask.height) {\n                const trimmed = new Uint8Array(mw * mh);\n                for (let row = 0; row < mh; row++) {\n                    trimmed.set(\n                        mask.data.subarray(row * mask.width, row * mask.width + mw),\n                        row * mw,\n                    );\n                }\n                finalMask = new Mask(trimmed, mw, mh);\n            }\n        } else {\n            finalMask = new Mask(new Uint8Array(0), 0, 0);\n            segmentedImage = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            mask: finalMask,\n            segmentedImage,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly SegmentationResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as SegmentationResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n\n    private _buildMasks(\n        detections: readonly SegmentationResult[],\n        origShape: readonly [number, number],\n    ): Masks {\n        const xyxy = new Float32Array(detections.length * 4);\n        for (let i = 0; i < detections.length; i++) {\n            const d = detections[i] as SegmentationResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n        }\n        return new Masks(\n            detections.map((d) => d.mask),\n            xyxy,\n            origShape,\n        );\n    }\n}\n"],"mappings":"+RAqFA,IAAa,EAAb,MAAa,UAAkB,EAAA,UAAW,CAGjB,MACA,QACA,OACA,WACA,eACA,cACA,eACA,eATrB,YACI,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACF,CACE,MAAM,CAAO,EATI,KAAA,MAAA,EACA,KAAA,QAAA,EACA,KAAA,OAAA,EACA,KAAA,WAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,eAAA,EACA,KAAA,eAAA,CAGrB,CAGA,aAAa,OAAO,EAAoB,EAA4B,CAAC,EAAuB,CACxF,IAAM,EAAsB,EAAQ,MAAQ,WAC5C,GAAI,IAAS,WACT,MAAU,MAAM,+BAA+B,EAAK,0BAA0B,EAElF,IAAM,EAAU,MAAM,EAAA,WAAW,OAAO,EAAO,CAAO,EAChD,EAAS,EAAA,cAAc,EAAQ,QAAU,OAAQ,CACnD,WAAY,EAAQ,UACxB,CAAC,EACK,EAAgC,CAAC,EACvC,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAM,GAAK,EAAO,GAEtB,OAAO,IAAI,EACP,EACA,EACA,EACA,EACA,EAAQ,WAAa,CAAC,IAAK,GAAG,EAC9B,EAAQ,eAAiB,IACzB,EAAQ,cAAgB,IACxB,EAAQ,eAAiB,IACzB,EAAQ,eAAiB,EAC7B,CACJ,CAGA,IAAI,MAAsB,CACtB,OAAO,KAAK,KAChB,CAGA,IAAI,QAA4B,CAC5B,OAAO,KAAK,OAChB,CAGA,IAAI,OAA0C,CAC1C,OAAO,KAAK,MAChB,CAGA,IAAI,YAAqB,CACrB,OAAO,KAAK,QAAQ,MACxB,CAGA,MAAM,KACF,EACA,EAAmC,CAAC,EACN,CAC9B,OAAO,KAAK,QAAQ,EAAO,CAAO,CACtC,CAGA,MAAM,QACF,EACA,EAAmC,CAAC,EACN,CAC9B,IAAM,EAAQ,IAAI,EAAA,WACZ,EAAO,OAAO,GAAU,SAAW,EAAQ,KAC3C,EAAW,MAAM,EAAA,UAAU,CAAK,EACtC,EAAM,MAAM,MAAM,EAClB,GAAM,CAAE,SAAQ,QAAO,UAAS,UAAW,KAAK,YAAY,CAAQ,EACpE,EAAM,MAAM,YAAY,EACxB,IAAM,EAAU,MAAM,KAAK,SAAS,IAAI,EAAG,KAAK,SAAS,WAAY,CAAO,CAAC,EAC7E,EAAM,MAAM,WAAW,EAEvB,GAAM,CAAE,YAAW,cAAe,KAAK,cAAc,CAAO,EAEtD,EAAa,EAAA,cACf,EAAU,KACV,EAAU,KACV,EAAW,KACX,EAAW,KACX,CACI,WAAY,KAAK,QAAQ,OACzB,WAAY,KAAK,WAAW,GAC5B,YAAa,KAAK,WAAW,GAC7B,cAAe,EAAS,MACxB,eAAgB,EAAS,OACzB,UACA,SACA,QACA,cAAe,EAAQ,eAAiB,KAAK,eAC7C,aAAc,EAAQ,cAAgB,KAAK,cAC3C,cAAe,KAAK,eACpB,cAAe,KAAK,cACxB,CACJ,EAUM,GAPF,EAAQ,UAAY,IAAA,GAKd,OAJO,CACH,IAAM,EAAU,IAAI,IAAI,EAAQ,OAAO,EACvC,OAAO,EAAW,OAAQ,GAAM,EAAQ,IAAI,EAAE,OAAO,CAAC,CAC1D,EAAA,CAAG,EACH,CAEiB,IAAK,GAC5B,KAAK,aAAa,EAAU,EAAE,KAAM,EAAE,QAAS,EAAE,WAAY,EAAE,IAAI,CACvE,EAEM,EAAkC,CAAC,EAAS,OAAQ,EAAS,KAAK,EAClE,EAAQ,KAAK,YAAY,EAAY,CAAI,EACzC,EAAQ,KAAK,YAAY,EAAY,CAAI,EAE/C,OADA,EAAM,MAAM,aAAa,EAClB,CACH,IAAI,EAAA,oBACA,EACA,EACA,EACA,KAAK,OACL,EACA,EACA,EACA,EAAM,MAAM,CAChB,CACJ,CACJ,CAEA,YAAoB,EAKlB,CACE,GAAM,CAAC,EAAI,GAAM,KAAK,WAChB,EAAK,EAAA,UAAU,EAAO,EAAI,CAAE,EAGlC,MAAO,CACH,OAAQ,EAAA,gBAFA,EAAA,MADA,EAAA,UAAU,EAAG,KACP,EAAK,EAAG,MAAM,MAAO,EAAG,MAAM,OAAQ,CAE5B,EAAK,CAAC,EAAG,EAAG,EAAG,MAAM,OAAQ,EAAG,MAAM,KAAK,CAAC,EACpE,MAAO,EAAG,MACV,QAAS,EAAG,QACZ,OAAQ,EAAG,MACf,CACJ,CAEA,cAAsB,EAGpB,CACE,IAAI,EACA,EACJ,IAAK,IAAM,KAAQ,KAAK,SAAS,YAAa,CAC1C,IAAM,EAAI,EAAQ,GACd,IAAM,IAAA,KACN,EAAE,KAAK,SAAW,GAAK,IAAc,IAAA,GACrC,EAAY,EACL,EAAE,KAAK,SAAW,GAAK,IAAe,IAAA,KAC7C,EAAa,GAErB,CACA,GAAI,IAAc,IAAA,IAAa,IAAe,IAAA,GAAW,CACrD,IAAM,EAAS,KAAK,SAAS,YAAY,IACpC,GAAM,GAAG,EAAE,IAAI,KAAK,UAAU,EAAQ,EAAE,EAAE,MAAQ,CAAC,CAAC,GACzD,EACA,MAAU,MACN,uDAAuD,EAAO,KAAK,IAAI,EAAE,GAC7E,CACJ,CACA,MAAO,CAAE,YAAW,YAAW,CACnC,CAEA,aACI,EACA,EACA,EACA,EACA,EACkB,CAClB,GAAM,CAAC,EAAI,EAAI,EAAI,GAAM,EAAK,UAAU,EAClC,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAS,MAAO,CAAE,EACjC,EAAM,KAAK,IAAI,EAAS,OAAQ,CAAE,EAEpC,EACA,EAAY,EAChB,GAAI,EAAM,GAAO,EAAM,GAAO,EAAK,KAAK,OAAS,EAAG,CAChD,IAAM,EAAQ,EAAM,EACd,EAAQ,EAAM,EACd,EAAK,KAAK,IAAI,EAAK,MAAO,CAAK,EAC/B,EAAK,KAAK,IAAI,EAAK,OAAQ,CAAK,EAChC,EAAU,IAAI,WAAW,EAAK,EAAK,CAAC,EAC1C,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAAO,CAC/B,IAAM,IAAiB,EAAM,GAAO,EAAS,MAAQ,GAAO,EACtD,EAAe,EAAM,EAAK,EAC1B,EAAgB,EAAM,EAAK,MACjC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAExB,GADU,EAAK,KAAK,EAAgB,KAC1B,EAAG,CACT,IAAM,EAAI,EAAe,EAAM,EACzB,EAAI,EAAe,EAAM,EAC/B,EAAQ,GAAK,EAAS,KAAK,GAC3B,EAAQ,EAAI,GAAK,EAAS,KAAK,EAAI,GACnC,EAAQ,EAAI,GAAK,EAAS,KAAK,EAAI,EACvC,CAER,CAEA,GADA,EAAiB,IAAI,EAAA,SAAS,EAAS,EAAI,CAAE,EACzC,IAAO,EAAK,OAAS,IAAO,EAAK,OAAQ,CACzC,IAAM,EAAU,IAAI,WAAW,EAAK,CAAE,EACtC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IACxB,EAAQ,IACJ,EAAK,KAAK,SAAS,EAAM,EAAK,MAAO,EAAM,EAAK,MAAQ,CAAE,EAC1D,EAAM,CACV,EAEJ,EAAY,IAAI,EAAA,KAAK,EAAS,EAAI,CAAE,CACxC,CACJ,KACI,GAAY,IAAI,EAAA,KAAK,IAAI,WAAe,EAAG,CAAC,EAC5C,EAAiB,IAAI,EAAA,SAAS,IAAI,WAAe,EAAG,CAAC,EAGzD,IAAM,EAAY,KAAK,OAAO,IAAY,SAAS,IAEnD,MAAO,CACH,UACA,YACA,aACA,OACA,IAAK,EACL,KAAM,EACN,KAAM,EACN,IAAK,EACL,KAAM,EACN,gBACJ,CACJ,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAI,EAAW,OACf,EAAO,IAAI,aAAa,EAAI,CAAC,EAC7B,EAAM,IAAI,WAAW,CAAC,EACtB,EAAO,IAAI,aAAa,CAAC,EAC/B,IAAK,IAAI,EAAI,EAAG,EAAI,EAAG,IAAK,CACxB,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAI,GAAK,EAAE,QACX,EAAK,GAAK,EAAE,UAChB,CACA,OAAO,IAAI,EAAA,MAAM,EAAM,EAAK,EAAM,CAAS,CAC/C,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAO,IAAI,aAAa,EAAW,OAAS,CAAC,EACnD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAW,OAAQ,IAAK,CACxC,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,EAC7B,CACA,OAAO,IAAI,EAAA,MACP,EAAW,IAAK,GAAM,EAAE,IAAI,EAC5B,EACA,CACJ,CACJ,CACJ"}
+{"version":3,"file":"segmenter.cjs","names":[],"sources":["../../../src/vision/tasks/segmenter.ts"],"sourcesContent":["/**\n * Instance-segmentation task using YOLO seg ONNX models (v8-seg / v11-seg / ...).\n */\n\nimport type * as ort from \"onnxruntime-web\";\n\nimport { type ModelSource, type OrtSessionOptions, OrtSession } from \"../core/session\";\nimport { SpeedTimer } from \"../core/timing\";\n\nimport { type ImageInput, loadImage } from \"../io/image\";\nimport { detectionNumClasses, resolveInputSize } from \"../core/graph\";\nimport { modelNames } from \"../core/metadata\";\nimport { type LabelSpec, resolveLabels } from \"../labels\";\nimport { decodeYoloSeg } from \"../postprocess/segmentation\";\nimport { letterbox, toCHW, toFloat32, toFloat32Tensor } from \"../preprocess/image\";\nimport { Boxes, Masks, SegmentationResults } from \"../results\";\nimport { VisionTask } from \"./base\";\nimport { type BoundingBox, type SegmentationResult, Mask, RGBImage } from \"../types\";\n\n/**\n * Decoder family for the segmentation head.\n *\n * - `\"yolo-seg\"`: YOLO instance-segmentation head with two outputs —\n *   `[1, 4 + nc + nm, N]` per-anchor predictions plus `[1, nm, mh, mw]`\n *   prototype masks. Covers YOLOv8-seg, v11-seg, v26-seg.\n *\n * The SDK does **not** auto-detect this — the caller is responsible for\n * picking a head that matches their export.\n */\nexport type SegmenterHead = \"yolo-seg\";\n\nexport interface SegmenterOptions extends OrtSessionOptions {\n    /**\n     * Decoder family for the segmentation head. Default `\"yolo-seg\"` covers\n     * YOLOv8-seg/v11-seg/v26-seg.\n     */\n    readonly head?: SegmenterHead;\n    /** Class label spec — see {@link resolveLabels}. Defaults to the COCO 80-class preset. */\n    readonly labels?: LabelSpec;\n    /** Number of classes — used to validate the supplied labels. */\n    readonly numClasses?: number;\n    /**\n     * Model input `[width, height]` in pixels for letterboxing.\n     *\n     * Only used when the model's graph leaves its spatial axes dynamic: a graph\n     * that declares a static size always wins, since that is the only shape ONNX\n     * Runtime will accept. Defaults to `[640, 640]`.\n     */\n    readonly inputSize?: readonly [number, number];\n    /** Default minimum class score to keep a candidate. */\n    readonly confThreshold?: number;\n    /** Default IoU threshold for non-maximum suppression. */\n    readonly iouThreshold?: number;\n    /** Maximum number of instances per image. */\n    readonly maxDetections?: number;\n    /** Probability cutoff applied to soft masks. Defaults to `0.5`. */\n    readonly maskThreshold?: number;\n}\n\nexport interface SegmenterPredictOptions {\n    readonly confThreshold?: number;\n    readonly iouThreshold?: number;\n    /**\n     * If set, keep only instances whose `classId` is in this list.\n     * Mirrors Ultralytics' `model.predict(img, classes=[0, 16])`.\n     */\n    readonly classes?: readonly number[];\n}\n\n/**\n * Instance segmenter for YOLO seg ONNX models (v8-seg / v11-seg / ...).\n *\n * The model is expected to expose two outputs:\n *\n * 1. `output0`: `(1, 4 + numClasses + numMaskCoefs, numAnchors)` — per-anchor\n *    predictions (boxes, class scores, mask coefficients).\n * 2. `output1`: `(1, numMaskCoefs, maskH, maskW)` — prototype masks.\n *\n * `predict()` returns `Promise<SegmentationResults[]>` (length 1 for a\n * single image), mirroring Ultralytics' API. The envelope exposes:\n *\n * - `boxes`: bulk numpy view (`xyxy`, `xywh`, `xyxyn`, `xywhn`, `cls`, `conf`).\n * - `masks`: per-instance binary masks cropped to each box.\n * - per-instance {@link SegmentationResult} via iteration.\n *\n * @example\n * ```typescript\n * const seg = await Segmenter.create(\"/models/yolov8n-seg.onnx\");\n * const r = (await seg.predict(\"/images/street.jpg\"))[0];\n * for (const inst of r) {\n *   console.log(inst.cls, inst.conf, inst.box.xyxy);\n * }\n * ```\n */\nexport class Segmenter extends VisionTask {\n    private constructor(\n        session: OrtSession,\n        private readonly _head: SegmenterHead,\n        private readonly _labels: readonly string[],\n        private readonly _names: Readonly<Record<number, string>>,\n        private readonly _inputSize: readonly [number, number],\n        private readonly _confThreshold: number,\n        private readonly _iouThreshold: number,\n        private readonly _maxDetections: number,\n        private readonly _maskThreshold: number,\n    ) {\n        super(session);\n    }\n\n    /** Load the model and resolve labels. */\n    static async create(model: ModelSource, options: SegmenterOptions = {}): Promise<Segmenter> {\n        const head: SegmenterHead = options.head ?? \"yolo-seg\";\n        if (head !== \"yolo-seg\") {\n            throw new Error(`Unsupported segmenter head '${head}'. Supported: 'yolo-seg'.`);\n        }\n        const session = await OrtSession.create(model, options);\n        const labels = resolveLabels(options.labels ?? modelNames(session.metadata) ?? \"coco\", {\n            numClasses: options.numClasses ?? detectionNumClasses(session.outputShape) ?? undefined,\n        });\n        const names: Record<number, string> = {};\n        for (let i = 0; i < labels.length; i++) {\n            names[i] = labels[i] as string;\n        }\n        return new Segmenter(\n            session,\n            head,\n            labels,\n            names,\n            resolveInputSize({\n                graphShape: session.inputShape,\n                requested: options.inputSize,\n                fallback: [640, 640],\n            }),\n            options.confThreshold ?? 0.25,\n            options.iouThreshold ?? 0.45,\n            options.maxDetections ?? 300,\n            options.maskThreshold ?? 0.5,\n        );\n    }\n\n    /** The decoder family used to interpret the model's output. */\n    get head(): SegmenterHead {\n        return this._head;\n    }\n\n    /** Class labels indexed by class id. */\n    get labels(): readonly string[] {\n        return this._labels;\n    }\n\n    /** Class id → class name dict (matches Ultralytics' `model.names`). */\n    get names(): Readonly<Record<number, string>> {\n        return this._names;\n    }\n\n    /**\n     * The `[width, height]` this task preprocesses to.\n     *\n     * Resolved at creation time from the model's graph when it declares a static\n     * input, so reading it back tells you the resolution inference really runs at\n     * — not merely what was requested.\n     */\n    get inputSize(): readonly [number, number] {\n        return this._inputSize;\n    }\n\n    /** Number of classes the model predicts. */\n    get numClasses(): number {\n        return this._labels.length;\n    }\n\n    /** Alias for {@link predict} (parity with PyTorch `nn.Module.__call__`). */\n    async call(\n        image: ImageInput,\n        options: SegmenterPredictOptions = {},\n    ): Promise<SegmentationResults[]> {\n        return this.predict(image, options);\n    }\n\n    /** Run instance segmentation on a single image. */\n    async predict(\n        image: ImageInput,\n        options: SegmenterPredictOptions = {},\n    ): Promise<SegmentationResults[]> {\n        const timer = new SpeedTimer();\n        const path = typeof image === \"string\" ? image : null;\n        const original = await loadImage(image);\n        timer.stage(\"load\");\n        const { tensor, scale, padLeft, padTop } = this._preprocess(original);\n        timer.stage(\"preprocess\");\n        const outputs = await this._session.run({ [this._session.inputName]: tensor });\n        timer.stage(\"inference\");\n\n        const { perAnchor, prototypes } = this._splitOutputs(outputs);\n\n        const decodedAll = decodeYoloSeg(\n            perAnchor.data as Float32Array,\n            perAnchor.dims,\n            prototypes.data as Float32Array,\n            prototypes.dims,\n            {\n                numClasses: this._labels.length,\n                inputWidth: this._inputSize[0],\n                inputHeight: this._inputSize[1],\n                originalWidth: original.width,\n                originalHeight: original.height,\n                padLeft,\n                padTop,\n                scale,\n                confThreshold: options.confThreshold ?? this._confThreshold,\n                iouThreshold: options.iouThreshold ?? this._iouThreshold,\n                maxDetections: this._maxDetections,\n                maskThreshold: this._maskThreshold,\n            },\n        );\n\n        const decoded =\n            options.classes !== undefined\n                ? (() => {\n                      const allowed = new Set(options.classes);\n                      return decodedAll.filter((d) => allowed.has(d.classId));\n                  })()\n                : decodedAll;\n\n        const detections = decoded.map((d) =>\n            this._buildResult(original, d.bbox, d.classId, d.confidence, d.mask),\n        );\n\n        const orig: readonly [number, number] = [original.height, original.width];\n        const boxes = this._buildBoxes(detections, orig);\n        const masks = this._buildMasks(detections, orig);\n        timer.stage(\"postprocess\");\n        return [\n            new SegmentationResults(\n                boxes,\n                masks,\n                detections,\n                this._names,\n                original,\n                orig,\n                path,\n                timer.speed(),\n            ),\n        ];\n    }\n\n    private _preprocess(image: RGBImage): {\n        tensor: ort.Tensor;\n        scale: number;\n        padLeft: number;\n        padTop: number;\n    } {\n        const [tw, th] = this._inputSize;\n        const lb = letterbox(image, tw, th);\n        const f32 = toFloat32(lb.image);\n        const chw = toCHW(f32, lb.image.width, lb.image.height, 3);\n        return {\n            tensor: toFloat32Tensor(chw, [1, 3, lb.image.height, lb.image.width]),\n            scale: lb.scale,\n            padLeft: lb.padLeft,\n            padTop: lb.padTop,\n        };\n    }\n\n    private _splitOutputs(outputs: Record<string, ort.Tensor>): {\n        perAnchor: ort.Tensor;\n        prototypes: ort.Tensor;\n    } {\n        let perAnchor: ort.Tensor | undefined;\n        let prototypes: ort.Tensor | undefined;\n        for (const name of this._session.outputNames) {\n            const t = outputs[name];\n            if (t === undefined) continue;\n            if (t.dims.length === 3 && perAnchor === undefined) {\n                perAnchor = t;\n            } else if (t.dims.length === 4 && prototypes === undefined) {\n                prototypes = t;\n            }\n        }\n        if (perAnchor === undefined || prototypes === undefined) {\n            const shapes = this._session.outputNames.map(\n                (n) => `${n}: ${JSON.stringify(outputs[n]?.dims ?? [])}`,\n            );\n            throw new Error(\n                `Segmenter expected one 3-D and one 4-D output, got [${shapes.join(\", \")}].`,\n            );\n        }\n        return { perAnchor, prototypes };\n    }\n\n    private _buildResult(\n        original: RGBImage,\n        bbox: BoundingBox,\n        classId: number,\n        confidence: number,\n        mask: Mask,\n    ): SegmentationResult {\n        const [x1, y1, x2, y2] = bbox.asIntXyxy();\n        const cx1 = Math.max(0, x1);\n        const cy1 = Math.max(0, y1);\n        const cx2 = Math.min(original.width, x2);\n        const cy2 = Math.min(original.height, y2);\n\n        let segmentedImage: RGBImage;\n        let finalMask = mask;\n        if (cx2 > cx1 && cy2 > cy1 && mask.data.length > 0) {\n            const cropW = cx2 - cx1;\n            const cropH = cy2 - cy1;\n            const mw = Math.min(mask.width, cropW);\n            const mh = Math.min(mask.height, cropH);\n            const segData = new Uint8Array(mw * mh * 3);\n            for (let row = 0; row < mh; row++) {\n                const srcRowOffset = ((cy1 + row) * original.width + cx1) * 3;\n                const dstRowOffset = row * mw * 3;\n                const maskRowOffset = row * mask.width;\n                for (let col = 0; col < mw; col++) {\n                    const m = mask.data[maskRowOffset + col];\n                    if (m !== 0) {\n                        const s = srcRowOffset + col * 3;\n                        const d = dstRowOffset + col * 3;\n                        segData[d] = original.data[s];\n                        segData[d + 1] = original.data[s + 1];\n                        segData[d + 2] = original.data[s + 2];\n                    }\n                }\n            }\n            segmentedImage = new RGBImage(segData, mw, mh);\n            if (mw !== mask.width || mh !== mask.height) {\n                const trimmed = new Uint8Array(mw * mh);\n                for (let row = 0; row < mh; row++) {\n                    trimmed.set(\n                        mask.data.subarray(row * mask.width, row * mask.width + mw),\n                        row * mw,\n                    );\n                }\n                finalMask = new Mask(trimmed, mw, mh);\n            }\n        } else {\n            finalMask = new Mask(new Uint8Array(0), 0, 0);\n            segmentedImage = new RGBImage(new Uint8Array(0), 0, 0);\n        }\n\n        const className = this._names[classId] ?? `class_${classId}`;\n\n        return {\n            classId,\n            className,\n            confidence,\n            bbox,\n            cls: classId,\n            name: className,\n            conf: confidence,\n            box: bbox,\n            mask: finalMask,\n            segmentedImage,\n        };\n    }\n\n    private _buildBoxes(\n        detections: readonly SegmentationResult[],\n        origShape: readonly [number, number],\n    ): Boxes {\n        const n = detections.length;\n        const xyxy = new Float32Array(n * 4);\n        const cls = new Int32Array(n);\n        const conf = new Float32Array(n);\n        for (let i = 0; i < n; i++) {\n            const d = detections[i] as SegmentationResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n            cls[i] = d.classId;\n            conf[i] = d.confidence;\n        }\n        return new Boxes(xyxy, cls, conf, origShape);\n    }\n\n    private _buildMasks(\n        detections: readonly SegmentationResult[],\n        origShape: readonly [number, number],\n    ): Masks {\n        const xyxy = new Float32Array(detections.length * 4);\n        for (let i = 0; i < detections.length; i++) {\n            const d = detections[i] as SegmentationResult;\n            xyxy[i * 4] = d.bbox.x1;\n            xyxy[i * 4 + 1] = d.bbox.y1;\n            xyxy[i * 4 + 2] = d.bbox.x2;\n            xyxy[i * 4 + 3] = d.bbox.y2;\n        }\n        return new Masks(\n            detections.map((d) => d.mask),\n            xyxy,\n            origShape,\n        );\n    }\n}\n"],"mappings":"gWA8FA,IAAa,EAAb,MAAa,UAAkB,EAAA,UAAW,CAGjB,MACA,QACA,OACA,WACA,eACA,cACA,eACA,eATrB,YACI,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACA,EACF,CACE,MAAM,CAAO,EATI,KAAA,MAAA,EACA,KAAA,QAAA,EACA,KAAA,OAAA,EACA,KAAA,WAAA,EACA,KAAA,eAAA,EACA,KAAA,cAAA,EACA,KAAA,eAAA,EACA,KAAA,eAAA,CAGrB,CAGA,aAAa,OAAO,EAAoB,EAA4B,CAAC,EAAuB,CACxF,IAAM,EAAsB,EAAQ,MAAQ,WAC5C,GAAI,IAAS,WACT,MAAU,MAAM,+BAA+B,EAAK,0BAA0B,EAElF,IAAM,EAAU,MAAM,EAAA,WAAW,OAAO,EAAO,CAAO,EAChD,EAAS,EAAA,cAAc,EAAQ,QAAU,EAAA,WAAW,EAAQ,QAAQ,GAAK,OAAQ,CACnF,WAAY,EAAQ,YAAc,EAAA,oBAAoB,EAAQ,WAAW,GAAK,IAAA,EAClF,CAAC,EACK,EAAgC,CAAC,EACvC,IAAK,IAAI,EAAI,EAAG,EAAI,EAAO,OAAQ,IAC/B,EAAM,GAAK,EAAO,GAEtB,OAAO,IAAI,EACP,EACA,EACA,EACA,EACA,EAAA,iBAAiB,CACb,WAAY,EAAQ,WACpB,UAAW,EAAQ,UACnB,SAAU,CAAC,IAAK,GAAG,CACvB,CAAC,EACD,EAAQ,eAAiB,IACzB,EAAQ,cAAgB,IACxB,EAAQ,eAAiB,IACzB,EAAQ,eAAiB,EAC7B,CACJ,CAGA,IAAI,MAAsB,CACtB,OAAO,KAAK,KAChB,CAGA,IAAI,QAA4B,CAC5B,OAAO,KAAK,OAChB,CAGA,IAAI,OAA0C,CAC1C,OAAO,KAAK,MAChB,CASA,IAAI,WAAuC,CACvC,OAAO,KAAK,UAChB,CAGA,IAAI,YAAqB,CACrB,OAAO,KAAK,QAAQ,MACxB,CAGA,MAAM,KACF,EACA,EAAmC,CAAC,EACN,CAC9B,OAAO,KAAK,QAAQ,EAAO,CAAO,CACtC,CAGA,MAAM,QACF,EACA,EAAmC,CAAC,EACN,CAC9B,IAAM,EAAQ,IAAI,EAAA,WACZ,EAAO,OAAO,GAAU,SAAW,EAAQ,KAC3C,EAAW,MAAM,EAAA,UAAU,CAAK,EACtC,EAAM,MAAM,MAAM,EAClB,GAAM,CAAE,SAAQ,QAAO,UAAS,UAAW,KAAK,YAAY,CAAQ,EACpE,EAAM,MAAM,YAAY,EACxB,IAAM,EAAU,MAAM,KAAK,SAAS,IAAI,EAAG,KAAK,SAAS,WAAY,CAAO,CAAC,EAC7E,EAAM,MAAM,WAAW,EAEvB,GAAM,CAAE,YAAW,cAAe,KAAK,cAAc,CAAO,EAEtD,EAAa,EAAA,cACf,EAAU,KACV,EAAU,KACV,EAAW,KACX,EAAW,KACX,CACI,WAAY,KAAK,QAAQ,OACzB,WAAY,KAAK,WAAW,GAC5B,YAAa,KAAK,WAAW,GAC7B,cAAe,EAAS,MACxB,eAAgB,EAAS,OACzB,UACA,SACA,QACA,cAAe,EAAQ,eAAiB,KAAK,eAC7C,aAAc,EAAQ,cAAgB,KAAK,cAC3C,cAAe,KAAK,eACpB,cAAe,KAAK,cACxB,CACJ,EAUM,GAPF,EAAQ,UAAY,IAAA,GAKd,OAJO,CACH,IAAM,EAAU,IAAI,IAAI,EAAQ,OAAO,EACvC,OAAO,EAAW,OAAQ,GAAM,EAAQ,IAAI,EAAE,OAAO,CAAC,CAC1D,EAAA,CAAG,EACH,CAEiB,IAAK,GAC5B,KAAK,aAAa,EAAU,EAAE,KAAM,EAAE,QAAS,EAAE,WAAY,EAAE,IAAI,CACvE,EAEM,EAAkC,CAAC,EAAS,OAAQ,EAAS,KAAK,EAClE,EAAQ,KAAK,YAAY,EAAY,CAAI,EACzC,EAAQ,KAAK,YAAY,EAAY,CAAI,EAE/C,OADA,EAAM,MAAM,aAAa,EAClB,CACH,IAAI,EAAA,oBACA,EACA,EACA,EACA,KAAK,OACL,EACA,EACA,EACA,EAAM,MAAM,CAChB,CACJ,CACJ,CAEA,YAAoB,EAKlB,CACE,GAAM,CAAC,EAAI,GAAM,KAAK,WAChB,EAAK,EAAA,UAAU,EAAO,EAAI,CAAE,EAGlC,MAAO,CACH,OAAQ,EAAA,gBAFA,EAAA,MADA,EAAA,UAAU,EAAG,KACP,EAAK,EAAG,MAAM,MAAO,EAAG,MAAM,OAAQ,CAE5B,EAAK,CAAC,EAAG,EAAG,EAAG,MAAM,OAAQ,EAAG,MAAM,KAAK,CAAC,EACpE,MAAO,EAAG,MACV,QAAS,EAAG,QACZ,OAAQ,EAAG,MACf,CACJ,CAEA,cAAsB,EAGpB,CACE,IAAI,EACA,EACJ,IAAK,IAAM,KAAQ,KAAK,SAAS,YAAa,CAC1C,IAAM,EAAI,EAAQ,GACd,IAAM,IAAA,KACN,EAAE,KAAK,SAAW,GAAK,IAAc,IAAA,GACrC,EAAY,EACL,EAAE,KAAK,SAAW,GAAK,IAAe,IAAA,KAC7C,EAAa,GAErB,CACA,GAAI,IAAc,IAAA,IAAa,IAAe,IAAA,GAAW,CACrD,IAAM,EAAS,KAAK,SAAS,YAAY,IACpC,GAAM,GAAG,EAAE,IAAI,KAAK,UAAU,EAAQ,EAAE,EAAE,MAAQ,CAAC,CAAC,GACzD,EACA,MAAU,MACN,uDAAuD,EAAO,KAAK,IAAI,EAAE,GAC7E,CACJ,CACA,MAAO,CAAE,YAAW,YAAW,CACnC,CAEA,aACI,EACA,EACA,EACA,EACA,EACkB,CAClB,GAAM,CAAC,EAAI,EAAI,EAAI,GAAM,EAAK,UAAU,EAClC,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAG,CAAE,EACpB,EAAM,KAAK,IAAI,EAAS,MAAO,CAAE,EACjC,EAAM,KAAK,IAAI,EAAS,OAAQ,CAAE,EAEpC,EACA,EAAY,EAChB,GAAI,EAAM,GAAO,EAAM,GAAO,EAAK,KAAK,OAAS,EAAG,CAChD,IAAM,EAAQ,EAAM,EACd,EAAQ,EAAM,EACd,EAAK,KAAK,IAAI,EAAK,MAAO,CAAK,EAC/B,EAAK,KAAK,IAAI,EAAK,OAAQ,CAAK,EAChC,EAAU,IAAI,WAAW,EAAK,EAAK,CAAC,EAC1C,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAAO,CAC/B,IAAM,IAAiB,EAAM,GAAO,EAAS,MAAQ,GAAO,EACtD,EAAe,EAAM,EAAK,EAC1B,EAAgB,EAAM,EAAK,MACjC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IAExB,GADU,EAAK,KAAK,EAAgB,KAC1B,EAAG,CACT,IAAM,EAAI,EAAe,EAAM,EACzB,EAAI,EAAe,EAAM,EAC/B,EAAQ,GAAK,EAAS,KAAK,GAC3B,EAAQ,EAAI,GAAK,EAAS,KAAK,EAAI,GACnC,EAAQ,EAAI,GAAK,EAAS,KAAK,EAAI,EACvC,CAER,CAEA,GADA,EAAiB,IAAI,EAAA,SAAS,EAAS,EAAI,CAAE,EACzC,IAAO,EAAK,OAAS,IAAO,EAAK,OAAQ,CACzC,IAAM,EAAU,IAAI,WAAW,EAAK,CAAE,EACtC,IAAK,IAAI,EAAM,EAAG,EAAM,EAAI,IACxB,EAAQ,IACJ,EAAK,KAAK,SAAS,EAAM,EAAK,MAAO,EAAM,EAAK,MAAQ,CAAE,EAC1D,EAAM,CACV,EAEJ,EAAY,IAAI,EAAA,KAAK,EAAS,EAAI,CAAE,CACxC,CACJ,KACI,GAAY,IAAI,EAAA,KAAK,IAAI,WAAe,EAAG,CAAC,EAC5C,EAAiB,IAAI,EAAA,SAAS,IAAI,WAAe,EAAG,CAAC,EAGzD,IAAM,EAAY,KAAK,OAAO,IAAY,SAAS,IAEnD,MAAO,CACH,UACA,YACA,aACA,OACA,IAAK,EACL,KAAM,EACN,KAAM,EACN,IAAK,EACL,KAAM,EACN,gBACJ,CACJ,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAI,EAAW,OACf,EAAO,IAAI,aAAa,EAAI,CAAC,EAC7B,EAAM,IAAI,WAAW,CAAC,EACtB,EAAO,IAAI,aAAa,CAAC,EAC/B,IAAK,IAAI,EAAI,EAAG,EAAI,EAAG,IAAK,CACxB,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAI,GAAK,EAAE,QACX,EAAK,GAAK,EAAE,UAChB,CACA,OAAO,IAAI,EAAA,MAAM,EAAM,EAAK,EAAM,CAAS,CAC/C,CAEA,YACI,EACA,EACK,CACL,IAAM,EAAO,IAAI,aAAa,EAAW,OAAS,CAAC,EACnD,IAAK,IAAI,EAAI,EAAG,EAAI,EAAW,OAAQ,IAAK,CACxC,IAAM,EAAI,EAAW,GACrB,EAAK,EAAI,GAAK,EAAE,KAAK,GACrB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,GACzB,EAAK,EAAI,EAAI,GAAK,EAAE,KAAK,EAC7B,CACA,OAAO,IAAI,EAAA,MACP,EAAW,IAAK,GAAM,EAAE,IAAI,EAC5B,EACA,CACJ,CACJ,CACJ"}