@stabrise/scaledp 0.1.0

package/dist/detect-q8AI_Jdj.js

@@ -0,0 +1,274 @@
+import { h as getConfig, i as Stage, s as DetectionError } from "./pipeline-DACqGkpN.js";
+import { _ as boxFromBBox, m as toNchwFloat32, p as toImageData, s as decodeImage, u as letterbox } from "./image-CAH2rLv9.js";
+import { n as ensureModelFiles } from "./model-cache-BEaqqRZ9.js";
+import { i as resolveParams, t as BASE_STAGE_DEFAULTS } from "./params-DapwK9Ns.js";
+import { t as createDetectorOutput } from "./detector-output-lyF1Mqb8.js";
+import { t as createSession } from "./ort-CXDoPrtw.js";
+//#region src/detect/yolo-onnx.ts
+/**
+* YOLO object detection on onnxruntime-web.
+*
+* Mirrors `scaledp/models/detectors/YoloOnnxDetector.py`. Note the
+* preprocessing differs from the text detector: YOLO letterboxes with
+* *centred* padding and no mean/std normalisation, where DBNet pads bottom and
+* right and applies ImageNet statistics. Centred padding means the pad offsets
+* must be subtracted before unscaling.
+*/
+/** Fallback when the graph declares a symbolic input size. */
+const DEFAULT_YOLO_INPUT = 960;
+const YOLO_DETECTOR_DEFAULTS = Object.freeze({
+	...BASE_STAGE_DEFAULTS,
+	inputCol: "image",
+	outputCol: "boxes",
+	keepInputData: true,
+	model: "",
+	labels: [],
+	scoreThreshold: .2,
+	iouThreshold: .5,
+	padding: 0,
+	outputType: "yolo"
+});
+/** Axis-aligned IoU over xyxy boxes. */
+function iou(a, b) {
+	const x0 = Math.max(a[0], b[0]);
+	const y0 = Math.max(a[1], b[1]);
+	const x1 = Math.min(a[2], b[2]);
+	const y1 = Math.min(a[3], b[3]);
+	if (x1 <= x0 || y1 <= y0) return 0;
+	const intersection = (x1 - x0) * (y1 - y0);
+	const union = (a[2] - a[0]) * (a[3] - a[1]) + (b[2] - b[0]) * (b[3] - b[1]) - intersection;
+	return union <= 0 ? 0 : intersection / union;
+}
+/** Per-class non-maximum suppression; different classes never suppress each other. */
+function nonMaximumSuppression(detections, iouThreshold) {
+	const byScore = [...detections].sort((a, b) => b.score - a.score);
+	const kept = [];
+	for (const detection of byScore) if (!kept.some((other) => other.classId === detection.classId && iou(other.bbox, detection.bbox) > iouThreshold)) kept.push(detection);
+	return kept;
+}
+/**
+* Decode a YOLO output tensor.
+*
+* Two layouts are handled, because exports differ:
+*  - [1, 4 + numClasses, numAnchors] -- the classic v8/v11 transposed form,
+*    with cx,cy,w,h then per-class scores.
+*  - [1, numDetections, 6] -- an end-to-end NMS export: x0,y0,x1,y1,score,class.
+*/
+function decodeYoloOutput(data, dims, scoreThreshold) {
+	const out = [];
+	if (dims.length === 3 && dims[2] === 6) {
+		const count = dims[1];
+		for (let i = 0; i < count; i++) {
+			const o = i * 6;
+			const score = data[o + 4];
+			if (score < scoreThreshold) continue;
+			out.push({
+				bbox: [
+					data[o],
+					data[o + 1],
+					data[o + 2],
+					data[o + 3]
+				],
+				score,
+				classId: Math.round(data[o + 5])
+			});
+		}
+		return out;
+	}
+	if (dims.length !== 3) throw new DetectionError(`Unsupported YOLO output shape [${dims.join(", ")}]`, "decodeYoloOutput");
+	const channels = dims[1];
+	const anchors = dims[2];
+	const classCount = channels - 4;
+	for (let a = 0; a < anchors; a++) {
+		let best = -1;
+		let bestScore = 0;
+		for (let c = 0; c < classCount; c++) {
+			const score = data[(4 + c) * anchors + a];
+			if (score > bestScore) {
+				bestScore = score;
+				best = c;
+			}
+		}
+		if (best < 0 || bestScore < scoreThreshold) continue;
+		const cx = data[a];
+		const cy = data[anchors + a];
+		const w = data[2 * anchors + a];
+		const h = data[3 * anchors + a];
+		out.push({
+			bbox: [
+				cx - w / 2,
+				cy - h / 2,
+				cx + w / 2,
+				cy + h / 2
+			],
+			score: bestScore,
+			classId: best
+		});
+	}
+	return out;
+}
+var YoloOnnxDetector = class extends Stage {
+	name = "YoloOnnxDetector";
+	session = null;
+	loading = null;
+	inputSize = 960;
+	constructor(options = {}) {
+		super(resolveParams(YOLO_DETECTOR_DEFAULTS, options, { model: (value) => {
+			if (!value) throw new RangeError("model is required");
+		} }));
+	}
+	async init() {
+		await this.getSession();
+	}
+	getSession() {
+		if (this.session) return Promise.resolve(this.session);
+		if (this.loading) return this.loading;
+		this.loading = (async () => {
+			const { model } = this.params;
+			const spec = /^https?:\/\//.test(model) ? {
+				repo: "yolo",
+				files: [{ path: model }]
+			} : {
+				repo: model,
+				files: [{ path: "model.onnx" }]
+			};
+			const bytes = (await ensureModelFiles(spec))[spec.files[0]?.path ?? ""];
+			if (!bytes) throw new DetectionError(`Model ${model} not found`, this.name);
+			const session = await createSession(bytes, { executionProviders: getConfig().executionProviders });
+			this.inputSize = readInputSize(session);
+			this.session = session;
+			return session;
+		})();
+		this.loading.catch(() => {
+			this.loading = null;
+		});
+		return this.loading;
+	}
+	async dispose() {
+		await this.session?.release();
+		this.session = null;
+		this.loading = null;
+	}
+	async apply(input, row) {
+		const image = input;
+		if (image?.exception) throw new DetectionError(`Upstream stage failed: ${image.exception}`, this.name);
+		if (!image || !(image.data instanceof Uint8Array) || image.data.byteLength === 0) throw new DetectionError("Expected an Image with decoded bytes", this.name);
+		const bitmap = await decodeImage(image.data);
+		let detections;
+		try {
+			detections = await this.detect(bitmap);
+		} finally {
+			bitmap.close();
+		}
+		const bboxes = detections.map((d) => boxFromBBox(d.bbox, {
+			text: d.label,
+			score: d.score
+		}));
+		return createDetectorOutput({
+			path: String(row[this.params.pathCol] ?? "memory"),
+			type: this.params.outputType,
+			bboxes
+		});
+	}
+	/** Run the model over one decoded image, returning source-space detections. */
+	async detect(source) {
+		const session = await this.getSession();
+		const size = this.inputSize;
+		const fitted = letterbox(source, {
+			width: size,
+			height: size
+		}, {
+			padding: "center",
+			fill: "#ffffff"
+		});
+		const tensorData = toNchwFloat32(toImageData(fitted.canvas));
+		const { Tensor } = await import("onnxruntime-web");
+		const inputName = session.inputNames[0];
+		const outputName = session.outputNames[0];
+		if (!inputName || !outputName) throw new DetectionError("Model exposes no input or output", this.name);
+		const output = (await session.run({ [inputName]: new Tensor("float32", tensorData, [
+			1,
+			3,
+			size,
+			size
+		]) }))[outputName];
+		if (!output) throw new DetectionError(`Model produced no "${outputName}" output`, this.name);
+		const raw = decodeYoloOutput(output.data, output.dims, this.params.scoreThreshold);
+		const padX = Math.trunc((size - fitted.resized.width) / 2);
+		const padY = Math.trunc((size - fitted.resized.height) / 2);
+		const { labels, padding } = this.params;
+		return nonMaximumSuppression(raw.map(({ bbox, score, classId }) => {
+			const x0 = (bbox[0] - padX) / fitted.scale;
+			const y0 = (bbox[1] - padY) / fitted.scale;
+			const x1 = (bbox[2] - padX) / fitted.scale;
+			const y1 = (bbox[3] - padY) / fitted.scale;
+			const padW = (x1 - x0) * padding;
+			const padH = (y1 - y0) * padding;
+			return {
+				bbox: [
+					Math.max(0, x0 - padW),
+					Math.max(0, y0 - padH),
+					Math.min(source.width, x1 + padW),
+					Math.min(source.height, y1 + padH)
+				],
+				score,
+				classId,
+				label: labels[classId] ?? `class_${classId}`
+			};
+		}), this.params.iouThreshold);
+	}
+	onError(message, row) {
+		return createDetectorOutput({
+			path: String(row[this.params.pathCol] ?? "memory"),
+			type: this.params.outputType,
+			exception: message
+		});
+	}
+};
+/** Input side length from the graph, falling back when the dim is symbolic. */
+function readInputSize(session) {
+	const height = ((session.inputMetadata?.[0])?.dims)?.[2];
+	return typeof height === "number" && height > 0 ? height : 960;
+}
+//#endregion
+//#region src/detect/index.ts
+/**
+* Object detection for @stabrise/scaledp.
+*
+* Mirrors ScaleDP's YOLO detector stages. Requires the optional peer dependency
+* `onnxruntime-web`.
+*/
+/** Signature detection, mirroring ScaleDP's `SignatureDetector`. */
+var SignatureDetector = class extends YoloOnnxDetector {
+	name = "SignatureDetector";
+	constructor(options = {}) {
+		super({
+			...YOLO_DETECTOR_DEFAULTS,
+			model: "StabRise/signature_detection",
+			labels: ["signature"],
+			outputCol: "signatures",
+			outputType: "signature",
+			scoreThreshold: .2,
+			...options
+		});
+	}
+};
+/** Face detection, mirroring ScaleDP's `FaceDetector`. */
+var FaceDetector = class extends YoloOnnxDetector {
+	name = "FaceDetector";
+	constructor(options = {}) {
+		super({
+			...YOLO_DETECTOR_DEFAULTS,
+			model: "StabRise/face_detection",
+			labels: ["face"],
+			outputCol: "faces",
+			outputType: "face",
+			scoreThreshold: .2,
+			...options
+		});
+	}
+};
+//#endregion
+export { YoloOnnxDetector as a, nonMaximumSuppression as c, YOLO_DETECTOR_DEFAULTS as i, SignatureDetector as n, decodeYoloOutput as o, DEFAULT_YOLO_INPUT as r, iou as s, FaceDetector as t };
+
+//# sourceMappingURL=detect-q8AI_Jdj.js.map

package/dist/detect-q8AI_Jdj.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"detect-q8AI_Jdj.js","names":[],"sources":["../src/detect/yolo-onnx.ts","../src/detect/index.ts"],"sourcesContent":["/**\n * YOLO object detection on onnxruntime-web.\n *\n * Mirrors `scaledp/models/detectors/YoloOnnxDetector.py`. Note the\n * preprocessing differs from the text detector: YOLO letterboxes with\n * *centred* padding and no mean/std normalisation, where DBNet pads bottom and\n * right and applies ImageNet statistics. Centred padding means the pad offsets\n * must be subtracted before unscaling.\n */\n\nimport { getConfig } from '../core/config.js'\nimport { DetectionError } from '../core/errors.js'\nimport { decodeImage, letterbox, toImageData, toNchwFloat32 } from '../core/image.js'\nimport { ensureModelFiles } from '../core/model-cache.js'\nimport { BASE_STAGE_DEFAULTS, type BaseStageParams, resolveParams } from '../core/params.js'\nimport { type Row, Stage } from '../core/pipeline.js'\nimport { createSession } from '../ocr/ort.js'\nimport { type Box, boxFromBBox } from '../schemas/box.js'\nimport { createDetectorOutput, type DetectorOutput } from '../schemas/detector-output.js'\nimport type { ScaleDpImage } from '../schemas/image.js'\n\n/** Fallback when the graph declares a symbolic input size. */\nexport const DEFAULT_YOLO_INPUT = 960\n\nexport interface YoloOnnxDetectorParams extends BaseStageParams {\n    model: string\n    /** Class index -> label. An empty list falls back to `class_<n>`. */\n    labels: readonly string[]\n    scoreThreshold: number\n    /** IoU above which two same-class boxes are considered duplicates. */\n    iouThreshold: number\n    /** Grow each box by this fraction of its size, to avoid clipping edges. */\n    padding: number\n    /** Emitted as DetectorOutput.type. */\n    outputType: string\n}\n\nexport const YOLO_DETECTOR_DEFAULTS: YoloOnnxDetectorParams = Object.freeze({\n    ...BASE_STAGE_DEFAULTS,\n    inputCol: 'image',\n    outputCol: 'boxes',\n    keepInputData: true,\n    model: '',\n    labels: [] as readonly string[],\n    scoreThreshold: 0.2,\n    iouThreshold: 0.5,\n    padding: 0,\n    outputType: 'yolo',\n})\n\nexport interface Detection {\n    /** [x0, y0, x1, y1] in source-image coordinates. */\n    bbox: [number, number, number, number]\n    score: number\n    classId: number\n    label: string\n}\n\n/** Axis-aligned IoU over xyxy boxes. */\nexport function iou(a: readonly number[], b: readonly number[]): number {\n    const x0 = Math.max(a[0] as number, b[0] as number)\n    const y0 = Math.max(a[1] as number, b[1] as number)\n    const x1 = Math.min(a[2] as number, b[2] as number)\n    const y1 = Math.min(a[3] as number, b[3] as number)\n    if (x1 <= x0 || y1 <= y0) return 0\n\n    const intersection = (x1 - x0) * (y1 - y0)\n    const areaA = ((a[2] as number) - (a[0] as number)) * ((a[3] as number) - (a[1] as number))\n    const areaB = ((b[2] as number) - (b[0] as number)) * ((b[3] as number) - (b[1] as number))\n    const union = areaA + areaB - intersection\n    return union <= 0 ? 0 : intersection / union\n}\n\n/** Per-class non-maximum suppression; different classes never suppress each other. */\nexport function nonMaximumSuppression(detections: readonly Detection[], iouThreshold: number): Detection[] {\n    const byScore = [...detections].sort((a, b) => b.score - a.score)\n    const kept: Detection[] = []\n    for (const detection of byScore) {\n        const suppressed = kept.some(\n            (other) => other.classId === detection.classId && iou(other.bbox, detection.bbox) > iouThreshold\n        )\n        if (!suppressed) kept.push(detection)\n    }\n    return kept\n}\n\n/**\n * Decode a YOLO output tensor.\n *\n * Two layouts are handled, because exports differ:\n *  - [1, 4 + numClasses, numAnchors] -- the classic v8/v11 transposed form,\n *    with cx,cy,w,h then per-class scores.\n *  - [1, numDetections, 6] -- an end-to-end NMS export: x0,y0,x1,y1,score,class.\n */\nexport function decodeYoloOutput(\n    data: Float32Array,\n    dims: readonly number[],\n    scoreThreshold: number\n): { bbox: [number, number, number, number]; score: number; classId: number }[] {\n    const out: { bbox: [number, number, number, number]; score: number; classId: number }[] = []\n\n    if (dims.length === 3 && dims[2] === 6) {\n        const count = dims[1] as number\n        for (let i = 0; i < count; i++) {\n            const o = i * 6\n            const score = data[o + 4] as number\n            if (score < scoreThreshold) continue\n            out.push({\n                bbox: [\n                    data[o] as number,\n                    data[o + 1] as number,\n                    data[o + 2] as number,\n                    data[o + 3] as number,\n                ],\n                score,\n                classId: Math.round(data[o + 5] as number),\n            })\n        }\n        return out\n    }\n\n    if (dims.length !== 3) {\n        throw new DetectionError(`Unsupported YOLO output shape [${dims.join(', ')}]`, 'decodeYoloOutput')\n    }\n\n    const channels = dims[1] as number\n    const anchors = dims[2] as number\n    const classCount = channels - 4\n\n    for (let a = 0; a < anchors; a++) {\n        let best = -1\n        let bestScore = 0\n        for (let c = 0; c < classCount; c++) {\n            const score = data[(4 + c) * anchors + a] as number\n            if (score > bestScore) {\n                bestScore = score\n                best = c\n            }\n        }\n        if (best < 0 || bestScore < scoreThreshold) continue\n\n        const cx = data[a] as number\n        const cy = data[anchors + a] as number\n        const w = data[2 * anchors + a] as number\n        const h = data[3 * anchors + a] as number\n        out.push({\n            bbox: [cx - w / 2, cy - h / 2, cx + w / 2, cy + h / 2],\n            score: bestScore,\n            classId: best,\n        })\n    }\n    return out\n}\n\nexport class YoloOnnxDetector extends Stage<YoloOnnxDetectorParams> {\n    // Annotated as string rather than inferred as a literal, so SignatureDetector\n    // and FaceDetector can narrow it to their own names.\n    readonly name: string = 'YoloOnnxDetector'\n\n    private session: import('onnxruntime-web').InferenceSession | null = null\n    private loading: Promise<import('onnxruntime-web').InferenceSession> | null = null\n    private inputSize = DEFAULT_YOLO_INPUT\n\n    constructor(options: Partial<YoloOnnxDetectorParams> = {}) {\n        super(\n            resolveParams(YOLO_DETECTOR_DEFAULTS, options, {\n                model: (value) => {\n                    if (!value) throw new RangeError('model is required')\n                },\n            })\n        )\n    }\n\n    override async init(): Promise<void> {\n        await this.getSession()\n    }\n\n    private getSession(): Promise<import('onnxruntime-web').InferenceSession> {\n        if (this.session) return Promise.resolve(this.session)\n        if (this.loading) return this.loading\n\n        this.loading = (async () => {\n            const { model } = this.params\n            const spec = /^https?:\\/\\//.test(model)\n                ? { repo: 'yolo', files: [{ path: model }] }\n                : { repo: model, files: [{ path: 'model.onnx' }] }\n\n            const files = await ensureModelFiles(spec)\n            const bytes = files[spec.files[0]?.path ?? '']\n            if (!bytes) throw new DetectionError(`Model ${model} not found`, this.name)\n\n            const session = await createSession(bytes, {\n                executionProviders: getConfig().executionProviders,\n            })\n            this.inputSize = readInputSize(session)\n            this.session = session\n            return session\n        })()\n\n        this.loading.catch(() => {\n            this.loading = null\n        })\n        return this.loading\n    }\n\n    override async dispose(): Promise<void> {\n        await this.session?.release()\n        this.session = null\n        this.loading = null\n    }\n\n    protected async apply(input: unknown, row: Row): Promise<DetectorOutput> {\n        const image = input as ScaleDpImage | undefined\n        // Check `exception` first. A failed upstream stage returns a well-formed but\n        // empty Image, so testing the bytes first would report \"no decoded bytes\" and\n        // bury the real cause.\n        if (image?.exception) {\n            throw new DetectionError(`Upstream stage failed: ${image.exception}`, this.name)\n        }\n        if (!image || !(image.data instanceof Uint8Array) || image.data.byteLength === 0) {\n            throw new DetectionError('Expected an Image with decoded bytes', this.name)\n        }\n\n        const bitmap = await decodeImage(image.data)\n        let detections: Detection[]\n        try {\n            detections = await this.detect(bitmap)\n        } finally {\n            bitmap.close()\n        }\n\n        const bboxes: Box[] = detections.map((d) => boxFromBBox(d.bbox, { text: d.label, score: d.score }))\n        return createDetectorOutput({\n            path: String(row[this.params.pathCol] ?? 'memory'),\n            type: this.params.outputType,\n            bboxes,\n        })\n    }\n\n    /** Run the model over one decoded image, returning source-space detections. */\n    async detect(source: ImageBitmap | OffscreenCanvas): Promise<Detection[]> {\n        const session = await this.getSession()\n        const size = this.inputSize\n\n        // Centred padding, no mean/std -- YOLO's convention, unlike DBNet's.\n        const fitted = letterbox(\n            source,\n            { width: size, height: size },\n            {\n                padding: 'center',\n                fill: '#ffffff',\n            }\n        )\n        const tensorData = toNchwFloat32(toImageData(fitted.canvas))\n\n        const { Tensor } = await import('onnxruntime-web')\n        const inputName = session.inputNames[0]\n        const outputName = session.outputNames[0]\n        if (!inputName || !outputName) {\n            throw new DetectionError('Model exposes no input or output', this.name)\n        }\n\n        const outputs = await session.run({\n            [inputName]: new Tensor('float32', tensorData, [1, 3, size, size]),\n        })\n        const output = outputs[outputName]\n        if (!output) throw new DetectionError(`Model produced no \"${outputName}\" output`, this.name)\n\n        const raw = decodeYoloOutput(output.data as Float32Array, output.dims, this.params.scoreThreshold)\n\n        const padX = Math.trunc((size - fitted.resized.width) / 2)\n        const padY = Math.trunc((size - fitted.resized.height) / 2)\n        const { labels, padding } = this.params\n\n        const detections: Detection[] = raw.map(({ bbox, score, classId }) => {\n            // Undo the centred pad first, then the scale.\n            const x0 = (bbox[0] - padX) / fitted.scale\n            const y0 = (bbox[1] - padY) / fitted.scale\n            const x1 = (bbox[2] - padX) / fitted.scale\n            const y1 = (bbox[3] - padY) / fitted.scale\n\n            const padW = (x1 - x0) * padding\n            const padH = (y1 - y0) * padding\n            return {\n                bbox: [\n                    Math.max(0, x0 - padW),\n                    Math.max(0, y0 - padH),\n                    Math.min(source.width, x1 + padW),\n                    Math.min(source.height, y1 + padH),\n                ],\n                score,\n                classId,\n                label: labels[classId] ?? `class_${classId}`,\n            }\n        })\n\n        return nonMaximumSuppression(detections, this.params.iouThreshold)\n    }\n\n    protected onError(message: string, row: Row): DetectorOutput {\n        return createDetectorOutput({\n            path: String(row[this.params.pathCol] ?? 'memory'),\n            type: this.params.outputType,\n            exception: message,\n        })\n    }\n}\n\n/** Input side length from the graph, falling back when the dim is symbolic. */\nfunction readInputSize(session: import('onnxruntime-web').InferenceSession): number {\n    const meta = (\n        session as unknown as {\n            inputMetadata?: { dims?: (number | string)[] }[]\n        }\n    ).inputMetadata?.[0]\n    const dims = meta?.dims\n    const height = dims?.[2]\n    return typeof height === 'number' && height > 0 ? height : DEFAULT_YOLO_INPUT\n}\n","/**\n * Object detection for @stabrise/scaledp.\n *\n * Mirrors ScaleDP's YOLO detector stages. Requires the optional peer dependency\n * `onnxruntime-web`.\n */\n\nimport { YOLO_DETECTOR_DEFAULTS, YoloOnnxDetector, type YoloOnnxDetectorParams } from './yolo-onnx.js'\n\nexport type { Detection, YoloOnnxDetectorParams } from './yolo-onnx.js'\nexport {\n    DEFAULT_YOLO_INPUT,\n    decodeYoloOutput,\n    iou,\n    nonMaximumSuppression,\n    YOLO_DETECTOR_DEFAULTS,\n    YoloOnnxDetector,\n} from './yolo-onnx.js'\n\n/** Signature detection, mirroring ScaleDP's `SignatureDetector`. */\nexport class SignatureDetector extends YoloOnnxDetector {\n    override readonly name = 'SignatureDetector'\n\n    constructor(options: Partial<YoloOnnxDetectorParams> = {}) {\n        super({\n            ...YOLO_DETECTOR_DEFAULTS,\n            model: 'StabRise/signature_detection',\n            labels: ['signature'],\n            outputCol: 'signatures',\n            outputType: 'signature',\n            scoreThreshold: 0.2,\n            ...options,\n        })\n    }\n}\n\n/** Face detection, mirroring ScaleDP's `FaceDetector`. */\nexport class FaceDetector extends YoloOnnxDetector {\n    override readonly name = 'FaceDetector'\n\n    constructor(options: Partial<YoloOnnxDetectorParams> = {}) {\n        super({\n            ...YOLO_DETECTOR_DEFAULTS,\n            model: 'StabRise/face_detection',\n            labels: ['face'],\n            outputCol: 'faces',\n            outputType: 'face',\n            scoreThreshold: 0.2,\n            ...options,\n        })\n    }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;AAsBA,MAAa,qBAAqB;AAelC,MAAa,yBAAiD,OAAO,OAAO;CACxE,GAAG;CACH,UAAU;CACV,WAAW;CACX,eAAe;CACf,OAAO;CACP,QAAQ,CAAC;CACT,gBAAgB;CAChB,cAAc;CACd,SAAS;CACT,YAAY;AAChB,CAAC;;AAWD,SAAgB,IAAI,GAAsB,GAA8B;CACpE,MAAM,KAAK,KAAK,IAAI,EAAE,IAAc,EAAE,EAAY;CAClD,MAAM,KAAK,KAAK,IAAI,EAAE,IAAc,EAAE,EAAY;CAClD,MAAM,KAAK,KAAK,IAAI,EAAE,IAAc,EAAE,EAAY;CAClD,MAAM,KAAK,KAAK,IAAI,EAAE,IAAc,EAAE,EAAY;CAClD,IAAI,MAAM,MAAM,MAAM,IAAI,OAAO;CAEjC,MAAM,gBAAgB,KAAK,OAAO,KAAK;CAGvC,MAAM,SAFU,EAAE,KAAiB,EAAE,OAAmB,EAAE,KAAiB,EAAE,OAC7D,EAAE,KAAiB,EAAE,OAAmB,EAAE,KAAiB,EAAE,MAC/C;CAC9B,OAAO,SAAS,IAAI,IAAI,eAAe;AAC3C;;AAGA,SAAgB,sBAAsB,YAAkC,cAAmC;CACvG,MAAM,UAAU,CAAC,GAAG,UAAU,CAAC,CAAC,MAAM,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;CAChE,MAAM,OAAoB,CAAC;CAC3B,KAAK,MAAM,aAAa,SAIpB,IAAI,CAHe,KAAK,MACnB,UAAU,MAAM,YAAY,UAAU,WAAW,IAAI,MAAM,MAAM,UAAU,IAAI,IAAI,YAE1E,GAAG,KAAK,KAAK,SAAS;CAExC,OAAO;AACX;;;;;;;;;AAUA,SAAgB,iBACZ,MACA,MACA,gBAC4E;CAC5E,MAAM,MAAoF,CAAC;CAE3F,IAAI,KAAK,WAAW,KAAK,KAAK,OAAO,GAAG;EACpC,MAAM,QAAQ,KAAK;EACnB,KAAK,IAAI,IAAI,GAAG,IAAI,OAAO,KAAK;GAC5B,MAAM,IAAI,IAAI;GACd,MAAM,QAAQ,KAAK,IAAI;GACvB,IAAI,QAAQ,gBAAgB;GAC5B,IAAI,KAAK;IACL,MAAM;KACF,KAAK;KACL,KAAK,IAAI;KACT,KAAK,IAAI;KACT,KAAK,IAAI;IACb;IACA;IACA,SAAS,KAAK,MAAM,KAAK,IAAI,EAAY;GAC7C,CAAC;EACL;EACA,OAAO;CACX;CAEA,IAAI,KAAK,WAAW,GAChB,MAAM,IAAI,eAAe,kCAAkC,KAAK,KAAK,IAAI,EAAE,IAAI,kBAAkB;CAGrG,MAAM,WAAW,KAAK;CACtB,MAAM,UAAU,KAAK;CACrB,MAAM,aAAa,WAAW;CAE9B,KAAK,IAAI,IAAI,GAAG,IAAI,SAAS,KAAK;EAC9B,IAAI,OAAO;EACX,IAAI,YAAY;EAChB,KAAK,IAAI,IAAI,GAAG,IAAI,YAAY,KAAK;GACjC,MAAM,QAAQ,MAAM,IAAI,KAAK,UAAU;GACvC,IAAI,QAAQ,WAAW;IACnB,YAAY;IACZ,OAAO;GACX;EACJ;EACA,IAAI,OAAO,KAAK,YAAY,gBAAgB;EAE5C,MAAM,KAAK,KAAK;EAChB,MAAM,KAAK,KAAK,UAAU;EAC1B,MAAM,IAAI,KAAK,IAAI,UAAU;EAC7B,MAAM,IAAI,KAAK,IAAI,UAAU;EAC7B,IAAI,KAAK;GACL,MAAM;IAAC,KAAK,IAAI;IAAG,KAAK,IAAI;IAAG,KAAK,IAAI;IAAG,KAAK,IAAI;GAAC;GACrD,OAAO;GACP,SAAS;EACb,CAAC;CACL;CACA,OAAO;AACX;AAEA,IAAa,mBAAb,cAAsC,MAA8B;CAGhE,OAAwB;CAExB,UAAqE;CACrE,UAA8E;CAC9E,YAAQ;CAER,YAAY,UAA2C,CAAC,GAAG;EACvD,MACI,cAAc,wBAAwB,SAAS,EAC3C,QAAQ,UAAU;GACd,IAAI,CAAC,OAAO,MAAM,IAAI,WAAW,mBAAmB;EACxD,EACJ,CAAC,CACL;CACJ;CAEA,MAAe,OAAsB;EACjC,MAAM,KAAK,WAAW;CAC1B;CAEA,aAA0E;EACtE,IAAI,KAAK,SAAS,OAAO,QAAQ,QAAQ,KAAK,OAAO;EACrD,IAAI,KAAK,SAAS,OAAO,KAAK;EAE9B,KAAK,WAAW,YAAY;GACxB,MAAM,EAAE,UAAU,KAAK;GACvB,MAAM,OAAO,eAAe,KAAK,KAAK,IAChC;IAAE,MAAM;IAAQ,OAAO,CAAC,EAAE,MAAM,MAAM,CAAC;GAAE,IACzC;IAAE,MAAM;IAAO,OAAO,CAAC,EAAE,MAAM,aAAa,CAAC;GAAE;GAGrD,MAAM,SAAQ,MADM,iBAAiB,IAAI,EAAA,CACrB,KAAK,MAAM,EAAE,EAAE,QAAQ;GAC3C,IAAI,CAAC,OAAO,MAAM,IAAI,eAAe,SAAS,MAAM,aAAa,KAAK,IAAI;GAE1E,MAAM,UAAU,MAAM,cAAc,OAAO,EACvC,oBAAoB,UAAU,CAAC,CAAC,mBACpC,CAAC;GACD,KAAK,YAAY,cAAc,OAAO;GACtC,KAAK,UAAU;GACf,OAAO;EACX,EAAA,CAAG;EAEH,KAAK,QAAQ,YAAY;GACrB,KAAK,UAAU;EACnB,CAAC;EACD,OAAO,KAAK;CAChB;CAEA,MAAe,UAAyB;EACpC,MAAM,KAAK,SAAS,QAAQ;EAC5B,KAAK,UAAU;EACf,KAAK,UAAU;CACnB;CAEA,MAAgB,MAAM,OAAgB,KAAmC;EACrE,MAAM,QAAQ;EAId,IAAI,OAAO,WACP,MAAM,IAAI,eAAe,0BAA0B,MAAM,aAAa,KAAK,IAAI;EAEnF,IAAI,CAAC,SAAS,EAAE,MAAM,gBAAgB,eAAe,MAAM,KAAK,eAAe,GAC3E,MAAM,IAAI,eAAe,wCAAwC,KAAK,IAAI;EAG9E,MAAM,SAAS,MAAM,YAAY,MAAM,IAAI;EAC3C,IAAI;EACJ,IAAI;GACA,aAAa,MAAM,KAAK,OAAO,MAAM;EACzC,UAAU;GACN,OAAO,MAAM;EACjB;EAEA,MAAM,SAAgB,WAAW,KAAK,MAAM,YAAY,EAAE,MAAM;GAAE,MAAM,EAAE;GAAO,OAAO,EAAE;EAAM,CAAC,CAAC;EAClG,OAAO,qBAAqB;GACxB,MAAM,OAAO,IAAI,KAAK,OAAO,YAAY,QAAQ;GACjD,MAAM,KAAK,OAAO;GAClB;EACJ,CAAC;CACL;;CAGA,MAAM,OAAO,QAA6D;EACtE,MAAM,UAAU,MAAM,KAAK,WAAW;EACtC,MAAM,OAAO,KAAK;EAGlB,MAAM,SAAS,UACX,QACA;GAAE,OAAO;GAAM,QAAQ;EAAK,GAC5B;GACI,SAAS;GACT,MAAM;EACV,CACJ;EACA,MAAM,aAAa,cAAc,YAAY,OAAO,MAAM,CAAC;EAE3D,MAAM,EAAE,WAAW,MAAM,OAAO;EAChC,MAAM,YAAY,QAAQ,WAAW;EACrC,MAAM,aAAa,QAAQ,YAAY;EACvC,IAAI,CAAC,aAAa,CAAC,YACf,MAAM,IAAI,eAAe,oCAAoC,KAAK,IAAI;EAM1E,MAAM,UAAS,MAHO,QAAQ,IAAI,GAC7B,YAAY,IAAI,OAAO,WAAW,YAAY;GAAC;GAAG;GAAG;GAAM;EAAI,CAAC,EACrE,CAAC,EAAA,CACsB;EACvB,IAAI,CAAC,QAAQ,MAAM,IAAI,eAAe,sBAAsB,WAAW,WAAW,KAAK,IAAI;EAE3F,MAAM,MAAM,iBAAiB,OAAO,MAAsB,OAAO,MAAM,KAAK,OAAO,cAAc;EAEjG,MAAM,OAAO,KAAK,OAAO,OAAO,OAAO,QAAQ,SAAS,CAAC;EACzD,MAAM,OAAO,KAAK,OAAO,OAAO,OAAO,QAAQ,UAAU,CAAC;EAC1D,MAAM,EAAE,QAAQ,YAAY,KAAK;EAwBjC,OAAO,sBAtByB,IAAI,KAAK,EAAE,MAAM,OAAO,cAAc;GAElE,MAAM,MAAM,KAAK,KAAK,QAAQ,OAAO;GACrC,MAAM,MAAM,KAAK,KAAK,QAAQ,OAAO;GACrC,MAAM,MAAM,KAAK,KAAK,QAAQ,OAAO;GACrC,MAAM,MAAM,KAAK,KAAK,QAAQ,OAAO;GAErC,MAAM,QAAQ,KAAK,MAAM;GACzB,MAAM,QAAQ,KAAK,MAAM;GACzB,OAAO;IACH,MAAM;KACF,KAAK,IAAI,GAAG,KAAK,IAAI;KACrB,KAAK,IAAI,GAAG,KAAK,IAAI;KACrB,KAAK,IAAI,OAAO,OAAO,KAAK,IAAI;KAChC,KAAK,IAAI,OAAO,QAAQ,KAAK,IAAI;IACrC;IACA;IACA;IACA,OAAO,OAAO,YAAY,SAAS;GACvC;EACJ,CAEsC,GAAG,KAAK,OAAO,YAAY;CACrE;CAEA,QAAkB,SAAiB,KAA0B;EACzD,OAAO,qBAAqB;GACxB,MAAM,OAAO,IAAI,KAAK,OAAO,YAAY,QAAQ;GACjD,MAAM,KAAK,OAAO;GAClB,WAAW;EACf,CAAC;CACL;AACJ;;AAGA,SAAS,cAAc,SAA6D;CAOhF,MAAM,WALF,QAGF,gBAAgB,GAAA,EACC,KAAA,GACG;CACtB,OAAO,OAAO,WAAW,YAAY,SAAS,IAAI,SAAA;AACtD;;;;;;;;;;AC1SA,IAAa,oBAAb,cAAuC,iBAAiB;CACpD,OAAyB;CAEzB,YAAY,UAA2C,CAAC,GAAG;EACvD,MAAM;GACF,GAAG;GACH,OAAO;GACP,QAAQ,CAAC,WAAW;GACpB,WAAW;GACX,YAAY;GACZ,gBAAgB;GAChB,GAAG;EACP,CAAC;CACL;AACJ;;AAGA,IAAa,eAAb,cAAkC,iBAAiB;CAC/C,OAAyB;CAEzB,YAAY,UAA2C,CAAC,GAAG;EACvD,MAAM;GACF,GAAG;GACH,OAAO;GACP,QAAQ,CAAC,MAAM;GACf,WAAW;GACX,YAAY;GACZ,gBAAgB;GAChB,GAAG;EACP,CAAC;CACL;AACJ"}

package/dist/detector-output-C0Qt-jEq.d.ts

@@ -0,0 +1,13 @@
+import { n as Box } from "./box-DAfzwfhA.js";
+//#region src/schemas/detector-output.d.ts
+interface DetectorOutput {
+  path: string;
+  /** Engine that produced the boxes: 'paddle' | 'dbnet-onnx' | 'yolo' | 'tesseract' | ... */
+  type: string;
+  bboxes: Box[];
+  exception: string;
+}
+declare function createDetectorOutput(init?: Partial<DetectorOutput>): DetectorOutput;
+//#endregion
+export { createDetectorOutput as n, DetectorOutput as t };
+//# sourceMappingURL=detector-output-C0Qt-jEq.d.ts.map

package/dist/detector-output-lyF1Mqb8.js

@@ -0,0 +1,13 @@
+//#region src/schemas/detector-output.ts
+function createDetectorOutput(init = {}) {
+	return {
+		path: init.path ?? "memory",
+		type: init.type ?? "detector",
+		bboxes: init.bboxes ?? [],
+		exception: init.exception ?? ""
+	};
+}
+//#endregion
+export { createDetectorOutput as t };
+
+//# sourceMappingURL=detector-output-lyF1Mqb8.js.map

package/dist/detector-output-lyF1Mqb8.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"detector-output-lyF1Mqb8.js","names":[],"sources":["../src/schemas/detector-output.ts"],"sourcesContent":["/** Port of `scaledp/schemas/DetectorOutput.py`. */\n\nimport type { Box } from './box.js'\n\nexport interface DetectorOutput {\n    path: string\n    /** Engine that produced the boxes: 'paddle' | 'dbnet-onnx' | 'yolo' | 'tesseract' | ... */\n    type: string\n    bboxes: Box[]\n    exception: string\n}\n\nexport function createDetectorOutput(init: Partial<DetectorOutput> = {}): DetectorOutput {\n    return {\n        path: init.path ?? 'memory',\n        type: init.type ?? 'detector',\n        bboxes: init.bboxes ?? [],\n        exception: init.exception ?? '',\n    }\n}\n"],"mappings":";AAYA,SAAgB,qBAAqB,OAAgC,CAAC,GAAmB;CACrF,OAAO;EACH,MAAM,KAAK,QAAQ;EACnB,MAAM,KAAK,QAAQ;EACnB,QAAQ,KAAK,UAAU,CAAC;EACxB,WAAW,KAAK,aAAa;CACjC;AACJ"}

package/dist/display/index.d.ts

@@ -0,0 +1,66 @@
+import { t as DetectorOutput } from "../detector-output-C0Qt-jEq.js";
+import { t as Document } from "../document-B8I61TiY.js";
+import { n as NerOutput, t as Entity } from "../entity-CedtRhU1.js";
+import { n as ScaleDpImage } from "../image-Dc5TSg46.js";
+//#region src/display/index.d.ts
+/**
+ * A stable colour per entity group.
+ *
+ * Python picks a random colour each run, so two renders of the same document
+ * never match. Hashing the group name keeps 'PERSON' one colour everywhere.
+ */
+declare function colorForGroup(name: string): string;
+/** Replace a container's contents with `node`. Accepts a selector or element. */
+declare function renderInto(target: string | HTMLElement, node: Node): HTMLElement;
+interface ShowImageOptions {
+  /** CSS width, e.g. '600px' or '100%'. */
+  width?: string;
+  alt?: string;
+}
+/**
+ * An `<img>` for a ScaleDP image. Mirrors `show_image`.
+ *
+ * The object URL is revoked once the image has decoded -- holding one per page
+ * leaks the whole blob for the lifetime of the document.
+ */
+declare function showImage(image: ScaleDpImage, options?: ShowImageOptions): HTMLElement;
+interface ShowTextOptions {
+  /**
+   * Preserve the document's own layout. Correct when the OCR stage ran with
+   * `keepFormatting`, which encodes the layout in spaces and blank lines.
+   */
+  preserveLayout?: boolean;
+  maxHeight?: string;
+}
+/** A `<pre>` of the recognized text. Mirrors `show_text`. */
+declare function showText(document_: Document, options?: ShowTextOptions): HTMLElement;
+/** Pretty-printed JSON. Mirrors `show_json`. */
+declare function showJson(value: unknown, indent?: number): HTMLElement;
+interface ShowNerOptions {
+  /** Maximum rows; 0 shows all. Python defaults to 20. */
+  limit?: number;
+  /** Only these groups. */
+  whiteList?: readonly string[];
+}
+/** A table of entities. Mirrors `show_ner`. */
+declare function showNer(ner: NerOutput, options?: ShowNerOptions): HTMLElement;
+interface VisualizeNerOptions {
+  /** Only highlight these groups. */
+  labelsList?: readonly string[];
+  /** Render the group name beside each highlight. */
+  showLabels?: boolean;
+}
+/**
+ * The document text with entities highlighted inline. Mirrors `visualize_ner`.
+ *
+ * Splices spans by character offset, which is exactly what `Entity.start`/`end`
+ * index. Overlapping entities are dropped rather than nested: the highest-
+ * scoring one wins, because two spans cannot occupy the same characters in a
+ * flat text run.
+ */
+declare function visualizeNer(document_: Document, ner: NerOutput, options?: VisualizeNerOptions): HTMLElement;
+/** A summary table of detected boxes. */
+declare function showBoxes(output: DetectorOutput | Document, limit?: number): HTMLElement;
+//#endregion
+export { type DetectorOutput, type Document, type Entity, type NerOutput, type ScaleDpImage, ShowImageOptions, ShowNerOptions, ShowTextOptions, VisualizeNerOptions, colorForGroup, renderInto, showBoxes, showImage, showJson, showNer, showText, visualizeNer };
+//# sourceMappingURL=index.d.ts.map

package/dist/display/index.js

@@ -0,0 +1,237 @@
+//#region src/display/index.ts
+/**
+* A stable colour per entity group.
+*
+* Python picks a random colour each run, so two renders of the same document
+* never match. Hashing the group name keeps 'PERSON' one colour everywhere.
+*/
+function colorForGroup(name) {
+	let hash = 0;
+	for (let i = 0; i < name.length; i++) hash = hash * 31 + name.charCodeAt(i) | 0;
+	return `hsl(${Math.abs(hash) % 360}, 70%, 45%)`;
+}
+function element(tag, props = {}) {
+	const node = document.createElement(tag);
+	if (props.text !== void 0) node.textContent = props.text;
+	if (props.className) node.className = props.className;
+	if (props.style) Object.assign(node.style, props.style);
+	return node;
+}
+/** Replace a container's contents with `node`. Accepts a selector or element. */
+function renderInto(target, node) {
+	const host = typeof target === "string" ? document.querySelector(target) : target;
+	if (!host) throw new Error(`No element matches ${String(target)}`);
+	host.replaceChildren(node);
+	return host;
+}
+/**
+* An `<img>` for a ScaleDP image. Mirrors `show_image`.
+*
+* The object URL is revoked once the image has decoded -- holding one per page
+* leaks the whole blob for the lifetime of the document.
+*/
+function showImage(image, options = {}) {
+	if (image.exception) return errorBlock(image.exception);
+	const bytes = new Uint8Array(image.data.byteLength);
+	bytes.set(image.data);
+	const url = URL.createObjectURL(new Blob([bytes.buffer], { type: `image/${image.imageType}` }));
+	const img = element("img", { style: {
+		maxWidth: options.width ?? "100%",
+		height: "auto"
+	} });
+	img.alt = options.alt ?? image.path;
+	img.addEventListener("load", () => URL.revokeObjectURL(url), { once: true });
+	img.addEventListener("error", () => URL.revokeObjectURL(url), { once: true });
+	img.src = url;
+	return img;
+}
+/** A `<pre>` of the recognized text. Mirrors `show_text`. */
+function showText(document_, options = {}) {
+	if (document_.exception) return errorBlock(document_.exception);
+	return element("pre", {
+		text: document_.text,
+		className: "scaledp-text",
+		style: {
+			fontFamily: "ui-monospace, SFMono-Regular, Menlo, monospace",
+			fontSize: "12px",
+			whiteSpace: options.preserveLayout === false ? "pre-wrap" : "pre",
+			overflowX: "auto",
+			maxHeight: options.maxHeight ?? "30rem",
+			margin: "0"
+		}
+	});
+}
+/** Pretty-printed JSON. Mirrors `show_json`. */
+function showJson(value, indent = 2) {
+	return element("pre", {
+		text: typeof value === "string" ? value : JSON.stringify(value, null, indent),
+		className: "scaledp-json",
+		style: {
+			fontFamily: "ui-monospace, monospace",
+			fontSize: "12px",
+			overflowX: "auto"
+		}
+	});
+}
+/** A table of entities. Mirrors `show_ner`. */
+function showNer(ner, options = {}) {
+	if (ner.exception) return errorBlock(ner.exception);
+	const allowed = new Set(options.whiteList ?? []);
+	let entities = allowed.size > 0 ? ner.entities.filter((e) => allowed.has(e.entity_group)) : ner.entities;
+	const limit = options.limit ?? 20;
+	const total = entities.length;
+	if (limit > 0) entities = entities.slice(0, limit);
+	if (total === 0) return element("p", { text: "No entities found." });
+	const table = element("table", { className: "scaledp-ner" });
+	table.style.borderCollapse = "collapse";
+	const header = table.insertRow();
+	for (const label of [
+		"Type",
+		"Text",
+		"Score",
+		"Start",
+		"End",
+		"Boxes"
+	]) {
+		const th = document.createElement("th");
+		th.textContent = label;
+		th.style.cssText = "border:1px solid #ddd;padding:4px 8px;text-align:left";
+		header.append(th);
+	}
+	for (const entity of entities) {
+		const tr = table.insertRow();
+		const swatch = element("span", { style: {
+			display: "inline-block",
+			width: "8px",
+			height: "8px",
+			borderRadius: "50%",
+			marginRight: "6px",
+			background: colorForGroup(entity.entity_group)
+		} });
+		[
+			entity.entity_group,
+			entity.word,
+			entity.score.toFixed(3),
+			String(entity.start),
+			String(entity.end),
+			String(entity.boxes.length)
+		].forEach((value, index) => {
+			const cell = tr.insertCell();
+			cell.style.cssText = "border:1px solid #ddd;padding:4px 8px";
+			if (index === 0) cell.append(swatch);
+			cell.append(typeof value === "string" ? document.createTextNode(value) : value);
+		});
+	}
+	const wrapper = element("div");
+	wrapper.append(table);
+	if (limit > 0 && total > limit) wrapper.append(element("p", { text: `Showing ${limit} of ${total} entities.` }));
+	return wrapper;
+}
+/**
+* The document text with entities highlighted inline. Mirrors `visualize_ner`.
+*
+* Splices spans by character offset, which is exactly what `Entity.start`/`end`
+* index. Overlapping entities are dropped rather than nested: the highest-
+* scoring one wins, because two spans cannot occupy the same characters in a
+* flat text run.
+*/
+function visualizeNer(document_, ner, options = {}) {
+	if (document_.exception) return errorBlock(document_.exception);
+	if (ner.exception) return errorBlock(ner.exception);
+	const allowed = new Set(options.labelsList ?? []);
+	const entities = (allowed.size > 0 ? ner.entities.filter((e) => allowed.has(e.entity_group)) : ner.entities).filter((e) => e.start >= 0 && e.end > e.start && e.end <= document_.text.length).sort((a, b) => a.start - b.start || b.score - a.score);
+	const container = element("div", {
+		className: "scaledp-ner-text",
+		style: {
+			fontFamily: "ui-monospace, SFMono-Regular, Menlo, monospace",
+			fontSize: "12px",
+			whiteSpace: "pre-wrap",
+			lineHeight: "1.9"
+		}
+	});
+	let cursor = 0;
+	for (const entity of entities) {
+		if (entity.start < cursor) continue;
+		if (entity.start > cursor) container.append(document.createTextNode(document_.text.slice(cursor, entity.start)));
+		const color = colorForGroup(entity.entity_group);
+		const mark = element("span", {
+			text: document_.text.slice(entity.start, entity.end),
+			style: {
+				background: color,
+				color: "#fff",
+				borderRadius: "3px",
+				padding: "1px 3px"
+			}
+		});
+		mark.title = `${entity.entity_group} (${entity.score.toFixed(3)})`;
+		container.append(mark);
+		if (options.showLabels) container.append(element("sup", {
+			text: entity.entity_group,
+			style: {
+				color,
+				fontSize: "9px",
+				marginLeft: "2px"
+			}
+		}));
+		cursor = entity.end;
+	}
+	container.append(document.createTextNode(document_.text.slice(cursor)));
+	return container;
+}
+/** A summary table of detected boxes. */
+function showBoxes(output, limit = 20) {
+	if (output.exception) return errorBlock(output.exception);
+	const boxes = output.bboxes;
+	const table = element("table");
+	table.style.borderCollapse = "collapse";
+	const header = table.insertRow();
+	for (const label of [
+		"Text",
+		"Score",
+		"x",
+		"y",
+		"w",
+		"h",
+		"angle"
+	]) {
+		const th = document.createElement("th");
+		th.textContent = label;
+		th.style.cssText = "border:1px solid #ddd;padding:4px 8px;text-align:left";
+		header.append(th);
+	}
+	for (const box of limit > 0 ? boxes.slice(0, limit) : boxes) {
+		const tr = table.insertRow();
+		for (const value of [
+			box.text,
+			box.score.toFixed(3),
+			String(box.x),
+			String(box.y),
+			String(box.width),
+			String(box.height),
+			box.angle.toFixed(1)
+		]) {
+			const cell = tr.insertCell();
+			cell.style.cssText = "border:1px solid #ddd;padding:4px 8px";
+			cell.textContent = value;
+		}
+	}
+	const wrapper = element("div");
+	wrapper.append(table);
+	if (limit > 0 && boxes.length > limit) wrapper.append(element("p", { text: `Showing ${limit} of ${boxes.length} boxes.` }));
+	return wrapper;
+}
+function errorBlock(message) {
+	return element("pre", {
+		text: message,
+		style: {
+			color: "#b00020",
+			whiteSpace: "pre-wrap",
+			fontFamily: "ui-monospace, monospace",
+			fontSize: "12px"
+		}
+	});
+}
+//#endregion
+export { colorForGroup, renderInto, showBoxes, showImage, showJson, showNer, showText, visualizeNer };
+
+//# sourceMappingURL=index.js.map