@camstack/addon-pipeline 1.2.294 → 1.2.296

package/dist/{default-detection-model-D24AJOTn.js → default-detection-model-D1daTtqT.js}

@@ -1,4 +1,4 @@
-const require_dist = require("./dist-CJR259Xf.js");
+const require_dist = require("./dist-8up-f2TX.js");
 let node_crypto = require("node:crypto");
 //#region src/detection-pipeline/pipeline/landmark-precision-gate.ts
 /**
@@ -384,6 +384,24 @@ var OBJECT_DETECTION_MODELS = [
 			runtimes: ["python"]
 		} }
 	},
+	{
+		id: "ssdlite-mobiledet-coco-edgetpu",
+		name: "SSDLite MobileDet (Coral)",
+		description: "SSDLite MobileDet COCO (QAT) — EdgeTPU-compiled full-integer TFLite, 320×320; runs on the Coral USB Edge TPU. More accurate than SSD MobileNet V2 (32.9 vs 25.6 % COCO mAP, Coral-published). Scores top out near 0.77.",
+		inputSize: {
+			width: 320,
+			height: 320
+		},
+		labels: [],
+		preprocessMode: "letterbox",
+		postprocessor: "ssd",
+		license: "Apache-2.0",
+		formats: { tflite: {
+			url: hf("objectDetection/ssdlite-mobiledet/edgetpu/ssdlite_mobiledet_coco_qat_postprocess_edgetpu.tflite"),
+			sizeMB: 5.1,
+			runtimes: ["python"]
+		} }
+	},
 	ovPrecisionVariant("yolo26n", "objectDetection/yolo26/openvino", "YOLO26 Nano", "fp16", 5, true),
 	ovPrecisionVariant("yolo26n", "objectDetection/yolo26/openvino", "YOLO26 Nano", "int8", 3),
 	ovPrecisionVariant("yolo26s", "objectDetection/yolo26/openvino", "YOLO26 Small", "fp16", 19, true),
@@ -502,57 +520,90 @@ var OBJECT_DETECTION_MODELS = [
 * A retired id resolves to the step's format default instead.
 */
 var RETIRED_MODEL_IDS = { "face-detection": ["scrypted-yolov9t-face"] };
-var FACE_DETECTION_MODELS = [{
-	id: "scrfd-2.5g",
-	name: "SCRFD 2.5G",
-	description: "SCRFD 2.5G bnkps — balanced face detection with 5 keypoints (WIDER FACE 93.80/92.02/77.13). InsightFace pretrained weights.",
-	inputSize: {
-		width: 640,
-		height: 640
+var FACE_DETECTION_MODELS = [
+	{
+		id: "scrfd-2.5g",
+		name: "SCRFD 2.5G",
+		description: "SCRFD 2.5G bnkps — balanced face detection with 5 keypoints (WIDER FACE 93.80/92.02/77.13). InsightFace pretrained weights.",
+		inputSize: {
+			width: 640,
+			height: 640
+		},
+		labels: [{
+			id: "face",
+			name: "Face"
+		}],
+		preprocessMode: "letterbox",
+		inputNormalization: "scrfd",
+		license: "InsightFace-NonCommercial-Research",
+		formats: {
+			onnx: {
+				url: hf("faceDetection/scrfd/onnx/camstack-scrfd-2.5g.onnx"),
+				sizeMB: 3.1
+			},
+			coreml: {
+				url: hf("faceDetection/scrfd/coreml/camstack-scrfd-2.5g.mlpackage"),
+				sizeMB: 1.7,
+				isDirectory: true,
+				files: [...MLPACKAGE_FILES],
+				runtimes: ["python"]
+			},
+			openvino: ovFormat(hf("faceDetection/scrfd/openvino/camstack-scrfd-2.5g.xml"), 1.8)
+		}
 	},
-	labels: [{
-		id: "face",
-		name: "Face"
-	}],
-	preprocessMode: "letterbox",
-	inputNormalization: "scrfd",
-	license: "InsightFace-NonCommercial-Research",
-	formats: {
-		onnx: {
-			url: hf("faceDetection/scrfd/onnx/camstack-scrfd-2.5g.onnx"),
-			sizeMB: 3.1
+	{
+		id: "yunet-2023mar",
+		name: "YuNet 2023mar",
+		description: "YuNet (OpenCV Zoo, 2023mar) — tiny face detector with 5 keypoints, MIT weights (WIDER FACE AP 0.884/0.866/0.750). ~0.2 MB. Trained on WIDER FACE (CC BY-NC-ND images).",
+		inputSize: {
+			width: 640,
+			height: 640
 		},
-		coreml: {
-			url: hf("faceDetection/scrfd/coreml/camstack-scrfd-2.5g.mlpackage"),
-			sizeMB: 1.7,
-			isDirectory: true,
-			files: [...MLPACKAGE_FILES],
-			runtimes: ["python"]
+		labels: [{
+			id: "face",
+			name: "Face"
+		}],
+		preprocessMode: "letterbox",
+		postprocessor: "yunet",
+		license: "MIT",
+		formats: {
+			onnx: {
+				url: hf("faceDetection/yunet/onnx/camstack-yunet-2023mar.onnx"),
+				sizeMB: .23
+			},
+			coreml: {
+				url: hf("faceDetection/yunet/coreml/camstack-yunet-2023mar.mlpackage"),
+				sizeMB: .2,
+				isDirectory: true,
+				files: [...MLPACKAGE_FILES],
+				runtimes: ["python"]
+			},
+			openvino: ovFormat(hf("faceDetection/yunet/openvino/camstack-yunet-2023mar.xml"), .36)
+		}
+	},
+	{
+		id: "ssd-mobilenet-v2-face-edgetpu",
+		legacy: true,
+		name: "SSD MobileNet V2 Face (Coral)",
+		description: "SSD MobileNet V2 face detector — EdgeTPU-compiled full-integer TFLite, 320×320; runs on the Coral USB Edge TPU. Requires an ssd-face postprocessor (not yet available).",
+		inputSize: {
+			width: 320,
+			height: 320
 		},
-		openvino: ovFormat(hf("faceDetection/scrfd/openvino/camstack-scrfd-2.5g.xml"), 1.8)
+		labels: [{
+			id: "face",
+			name: "Face"
+		}],
+		preprocessMode: "letterbox",
+		postprocessor: "ssd",
+		license: "Apache-2.0",
+		formats: { tflite: {
+			url: "https://github.com/google-coral/test_data/raw/master/ssd_mobilenet_v2_face_quant_postprocess_edgetpu.tflite",
+			sizeMB: 6.7,
+			runtimes: ["python"]
+		} }
 	}
-}, {
-	id: "ssd-mobilenet-v2-face-edgetpu",
-	legacy: true,
-	name: "SSD MobileNet V2 Face (Coral)",
-	description: "SSD MobileNet V2 face detector — EdgeTPU-compiled full-integer TFLite, 320×320; runs on the Coral USB Edge TPU. Requires an ssd-face postprocessor (not yet available).",
-	inputSize: {
-		width: 320,
-		height: 320
-	},
-	labels: [{
-		id: "face",
-		name: "Face"
-	}],
-	preprocessMode: "letterbox",
-	postprocessor: "ssd",
-	license: "Apache-2.0",
-	formats: { tflite: {
-		url: "https://github.com/google-coral/test_data/raw/master/ssd_mobilenet_v2_face_quant_postprocess_edgetpu.tflite",
-		sizeMB: 6.7,
-		runtimes: ["python"]
-	} }
-}];
+];
 var FACE_EMBEDDING_MODELS = [
 	{
 		id: "arcface-r100",
@@ -1133,55 +1184,84 @@ var INSTANCE_SEGMENTATION_MODELS = [
 		}
 	}
 ];
-var CLIP_EMBEDDING_MODELS = [{
-	id: "mobileclip-s1",
-	name: "MobileCLIP S1",
-	description: "MobileCLIP S1 — Apple balanced CLIP vision encoder, 512-dim, 256×256 (fp16 OpenVINO/CoreML)",
-	inputSize: {
-		width: 256,
-		height: 256
+var CLIP_EMBEDDING_MODELS = [
+	{
+		id: "mobileclip-s1",
+		name: "MobileCLIP S1",
+		description: "MobileCLIP S1 — Apple balanced CLIP vision encoder, 512-dim, 256×256 (fp16 OpenVINO/CoreML)",
+		inputSize: {
+			width: 256,
+			height: 256
+		},
+		labels: [{
+			id: "embedding",
+			name: "CLIP Embedding"
+		}],
+		preprocessMode: "resize",
+		inputNormalization: "none",
+		formats: {
+			openvino: ovFormat(hf("clip/mobileclip-s1/openvino/camstack-mobileclip-s1-vision.xml"), 55),
+			coreml: {
+				url: hf("clip/mobileclip-s1/coreml/camstack-mobileclip-s1-vision.mlpackage"),
+				sizeMB: 65,
+				isDirectory: true,
+				files: [...MLPACKAGE_FILES],
+				runtimes: ["python"]
+			}
+		}
 	},
-	labels: [{
-		id: "embedding",
-		name: "CLIP Embedding"
-	}],
-	preprocessMode: "resize",
-	inputNormalization: "none",
-	formats: {
-		openvino: ovFormat(hf("clip/mobileclip-s1/openvino/camstack-mobileclip-s1-vision.xml"), 55),
-		coreml: {
-			url: hf("clip/mobileclip-s1/coreml/camstack-mobileclip-s1-vision.mlpackage"),
-			sizeMB: 65,
-			isDirectory: true,
-			files: [...MLPACKAGE_FILES],
-			runtimes: ["python"]
+	{
+		id: "mobileclip-s2",
+		name: "MobileCLIP S2",
+		description: "MobileCLIP S2 — Apple high-accuracy CLIP vision encoder, 512-dim, 256×256 (fp16 OpenVINO/CoreML)",
+		inputSize: {
+			width: 256,
+			height: 256
+		},
+		labels: [{
+			id: "embedding",
+			name: "CLIP Embedding"
+		}],
+		preprocessMode: "resize",
+		inputNormalization: "none",
+		formats: {
+			openvino: ovFormat(hf("clip/mobileclip-s2/openvino/camstack-mobileclip-s2-vision.xml"), 90),
+			coreml: {
+				url: hf("clip/mobileclip-s2/coreml/camstack-mobileclip-s2-vision.mlpackage"),
+				sizeMB: 110,
+				isDirectory: true,
+				files: [...MLPACKAGE_FILES],
+				runtimes: ["python"]
+			}
 		}
-	}
-}, {
-	id: "mobileclip-s2",
-	name: "MobileCLIP S2",
-	description: "MobileCLIP S2 — Apple high-accuracy CLIP vision encoder, 512-dim, 256×256 (fp16 OpenVINO/CoreML)",
-	inputSize: {
-		width: 256,
-		height: 256
 	},
-	labels: [{
-		id: "embedding",
-		name: "CLIP Embedding"
-	}],
-	preprocessMode: "resize",
-	inputNormalization: "none",
-	formats: {
-		openvino: ovFormat(hf("clip/mobileclip-s2/openvino/camstack-mobileclip-s2-vision.xml"), 90),
-		coreml: {
-			url: hf("clip/mobileclip-s2/coreml/camstack-mobileclip-s2-vision.mlpackage"),
-			sizeMB: 110,
-			isDirectory: true,
-			files: [...MLPACKAGE_FILES],
-			runtimes: ["python"]
+	{
+		id: "siglip2-b16-224",
+		name: "SigLIP2 B/16",
+		description: "Google SigLIP2 base, patch 16, 224×224 — Apache-2.0 CLIP vision encoder, 768-dim (fp16 OpenVINO/CoreML)",
+		inputSize: {
+			width: 224,
+			height: 224
+		},
+		labels: [{
+			id: "embedding",
+			name: "CLIP Embedding"
+		}],
+		preprocessMode: "resize",
+		inputNormalization: "none",
+		license: "Apache-2.0",
+		formats: {
+			openvino: ovFormat(hf("clip/siglip2/openvino/camstack-siglip2-b16-224-vision.xml"), 186),
+			coreml: {
+				url: hf("clip/siglip2/coreml/camstack-siglip2-b16-224-vision.mlpackage"),
+				sizeMB: 185,
+				isDirectory: true,
+				files: [...MLPACKAGE_FILES],
+				runtimes: ["python"]
+			}
 		}
 	}
-}];
+];
 var AUDIO_CLASSIFIER_MODELS = [{
 	id: "yamnet-onnx",
 	name: "YAMNet",
@@ -1574,7 +1654,8 @@ var STEP_FACE_DETECTION = new PipelineStepBase({
 	inputClasses: ["person"],
 	outputClasses: ["face"],
 	models: [...FACE_DETECTION_MODELS],
-	defaultModelId: "scrfd-2.5g",
+	defaultModelId: "yunet-2023mar",
+	defaultModelIdByFormat: { tflite: "scrfd-2.5g" },
 	defaultConfidence: .5,
 	defaultMinParentScore: .7,
 	cadence: {
@@ -1619,7 +1700,7 @@ var STEP_FACE_EMBEDDING = new FaceEmbeddingStep({
 	outputClasses: ["identity"],
 	labelTier: 2,
 	models: [...FACE_EMBEDDING_MODELS],
-	defaultModelId: "arcface-r100",
+	defaultModelId: "auraface-r100",
 	modelScope: "cluster",
 	defaultConfidence: 0,
 	cadence: {
@@ -1819,11 +1900,12 @@ function getStepDefinition(stepId) {
 *      has a build for `format`.
 *   2. `def.defaultModelId` — the step's plain declared default — if it
 *      exists in `def.models` AND has a build for `format`.
-*   3. The smallest-by-size model among those with a `format` build
-*      (legacy fallback, preserved for steps/formats with no declared
+*   3. The smallest-by-size NON-legacy model among those with a `format`
+*      build (fallback, preserved for steps/formats with no declared
 *      preference reachable).
-*   4. `def.defaultModelId` unchanged, when ZERO models have a `format`
-*      build — an unloadable case flagged elsewhere, not resolved here.
+*   4. When NO non-legacy model has a `format` build — an unloadable case
+*      flagged elsewhere, not resolved here — the declared per-format id if
+*      there is one, else `def.defaultModelId`, unchanged.
 */
 function getDefaultModelForFormat(stepId, format) {
 	return getDefaultModelForFormatFromDef(getStepDefinition(stepId), format);
@@ -1841,7 +1923,7 @@ function getDefaultModelForFormatFromDef(def, format) {
 	if (declaredForFormat !== void 0 && hasFormatBuild(declaredForFormat)) return declaredForFormat;
 	if (hasFormatBuild(def.defaultModelId)) return def.defaultModelId;
 	const available = def.models.filter((m) => m.formats[format] && m.legacy !== true);
-	if (available.length === 0) return def.defaultModelId;
+	if (available.length === 0) return declaredForFormat ?? def.defaultModelId;
 	return [...available].toSorted((a, b) => {
 		return (a.formats[format]?.sizeMB ?? Infinity) - (b.formats[format]?.sizeMB ?? Infinity);
 	})[0].id;