@camstack/addon-vision 0.1.0 → 0.1.2

package/dist/index.mjs

@@ -1,72 +1,241 @@
-import {
-  CameraNativeDetectionAddon
-} from "./chunk-DTOAB2CE.mjs";
 import {
   BirdGlobalClassifierAddon
-} from "./chunk-Q3SQOYG6.mjs";
+} from "./chunk-ZWYXXCXP.mjs";
 import {
   BirdNABirdsClassifierAddon
-} from "./chunk-3MQFUDRU.mjs";
+} from "./chunk-DRYFGARD.mjs";
 import {
   AnimalClassifierAddon
-} from "./chunk-J4WRYHHY.mjs";
+} from "./chunk-YUCD2TFH.mjs";
 import {
   ANIMAL_TYPE_MODELS,
   BIRD_NABIRDS_MODELS,
   BIRD_SPECIES_MODELS
-} from "./chunk-ISOIDU4U.mjs";
+} from "./chunk-DUN6XU3N.mjs";
+import {
+  VEHICLE_TYPE_MODELS
+} from "./chunk-XZ6ZMXXU.mjs";
+import {
+  SEGMENTATION_REFINER_MODELS
+} from "./chunk-7DYHXUPZ.mjs";
 import {
-  MotionDetectionAddon,
   detectMotion
-} from "./chunk-QIMDG34B.mjs";
+} from "./chunk-BR2FPGOX.mjs";
 import {
-  OBJECT_DETECTION_MODELS,
   ObjectDetectionAddon,
   SEGMENTATION_MODELS
-} from "./chunk-AYBFB7ID.mjs";
+} from "./chunk-R5J3WAUI.mjs";
 import {
   FACE_DETECTION_MODELS,
   FaceDetectionAddon,
   scrfdPostprocess
-} from "./chunk-MEVASN3P.mjs";
+} from "./chunk-ESLHNWWE.mjs";
 import {
   FACE_RECOGNITION_MODELS,
   FaceRecognitionAddon,
   cosineSimilarity,
   l2Normalize
-} from "./chunk-B3R66MPF.mjs";
+} from "./chunk-JUQEW6ON.mjs";
 import {
   PLATE_DETECTION_MODELS,
   PlateDetectionAddon
-} from "./chunk-5JJZGKL7.mjs";
+} from "./chunk-YPU4WTXZ.mjs";
 import {
   iou,
   nms,
   yoloPostprocess
 } from "./chunk-KUO2BVFY.mjs";
+import {
+  MLPACKAGE_FILES,
+  OBJECT_DETECTION_MODELS
+} from "./chunk-BP7H4NFS.mjs";
 import {
   PLATE_RECOGNITION_MODELS,
   PlateRecognitionAddon,
   ctcDecode
-} from "./chunk-PDSHDDPV.mjs";
+} from "./chunk-ZTJENCFC.mjs";
 import {
   cropRegion,
   jpegToRgb,
   letterbox,
   resizeAndNormalize,
   rgbToGrayscale
-} from "./chunk-6OR5TE7A.mjs";
+} from "./chunk-22BHCDT5.mjs";
 import {
   AUDIO_CLASSIFICATION_MODELS,
   AudioClassificationAddon,
   yamnetPostprocess
-} from "./chunk-5AIQSN32.mjs";
+} from "./chunk-D6WEHN33.mjs";
 import {
   NodeInferenceEngine,
   PythonInferenceEngine,
   probeOnnxBackends,
   resolveEngine
-} from "./chunk-LPI42WL6.mjs";
+} from "./chunk-2IOKI4ES.mjs";
+import "./chunk-BJTO5JO5.mjs";
+
+// src/catalogs/general-ocr-models.ts
+import { hfModelUrl } from "@camstack/types";
+var HF_REPO = "camstack/camstack-models";
+var OCR_TEXT_LABELS = [
+  { id: "text", name: "Scene Text" }
+];
+var GENERAL_OCR_MODELS = [
+  // ── OnnxTR / docTR — lightweight general scene text recognition ──
+  {
+    id: "doctr-det-db-mobilenet",
+    name: "docTR Detection MobileNet",
+    description: "docTR DBNet MobileNet V3 \u2014 lightweight text region detection for scene text",
+    inputSize: { width: 1024, height: 1024 },
+    labels: OCR_TEXT_LABELS,
+    formats: {
+      onnx: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/onnx/camstack-doctr-det-db-mobilenet-v3.onnx"),
+        sizeMB: 15
+      },
+      coreml: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/coreml/camstack-doctr-det-db-mobilenet-v3.mlpackage"),
+        sizeMB: 8,
+        isDirectory: true,
+        files: [
+          "Manifest.json",
+          "Data/com.apple.CoreML/model.mlmodel",
+          "Data/com.apple.CoreML/weights/weight.bin"
+        ],
+        runtimes: ["python"]
+      },
+      openvino: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/openvino/camstack-doctr-det-db-mobilenet-v3.xml"),
+        sizeMB: 8,
+        runtimes: ["python"]
+      }
+    }
+  },
+  {
+    id: "doctr-rec-crnn-mobilenet",
+    name: "docTR Recognition CRNN MobileNet",
+    description: "docTR CRNN MobileNet V3 \u2014 lightweight text recognition for detected regions",
+    inputSize: { width: 128, height: 32 },
+    labels: OCR_TEXT_LABELS,
+    formats: {
+      onnx: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/onnx/camstack-doctr-rec-crnn-mobilenet-v3.onnx"),
+        sizeMB: 5
+      },
+      coreml: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/coreml/camstack-doctr-rec-crnn-mobilenet-v3.mlpackage"),
+        sizeMB: 3,
+        isDirectory: true,
+        files: [
+          "Manifest.json",
+          "Data/com.apple.CoreML/model.mlmodel",
+          "Data/com.apple.CoreML/weights/weight.bin"
+        ],
+        runtimes: ["python"]
+      },
+      openvino: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/openvino/camstack-doctr-rec-crnn-mobilenet-v3.xml"),
+        sizeMB: 3,
+        runtimes: ["python"]
+      }
+    }
+  },
+  {
+    id: "doctr-rec-parseq",
+    name: "docTR Recognition PARSeq",
+    description: "docTR PARSeq \u2014 high-accuracy scene text recognition (top ICDAR scores)",
+    inputSize: { width: 128, height: 32 },
+    labels: OCR_TEXT_LABELS,
+    formats: {
+      onnx: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/onnx/camstack-doctr-rec-parseq.onnx"),
+        sizeMB: 25
+      },
+      coreml: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/coreml/camstack-doctr-rec-parseq.mlpackage"),
+        sizeMB: 13,
+        isDirectory: true,
+        files: [
+          "Manifest.json",
+          "Data/com.apple.CoreML/model.mlmodel",
+          "Data/com.apple.CoreML/weights/weight.bin"
+        ],
+        runtimes: ["python"]
+      },
+      openvino: {
+        url: hfModelUrl(HF_REPO, "generalOcr/doctr/openvino/camstack-doctr-rec-parseq.xml"),
+        sizeMB: 13,
+        runtimes: ["python"]
+      }
+    }
+  },
+  // ── PaddleOCR PP-OCRv5 Mobile — general-purpose text detection + recognition ──
+  {
+    id: "ppocr-v5-det-mobile",
+    name: "PP-OCRv5 Detection Mobile",
+    description: "PP-OCRv5 mobile text detection \u2014 optimized for edge, 100+ languages",
+    inputSize: { width: 640, height: 640 },
+    labels: OCR_TEXT_LABELS,
+    formats: {
+      onnx: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/onnx/camstack-ppocr-v5-det-mobile.onnx"),
+        sizeMB: 6
+      },
+      coreml: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/coreml/camstack-ppocr-v5-det-mobile.mlpackage"),
+        sizeMB: 3,
+        isDirectory: true,
+        files: [
+          "Manifest.json",
+          "Data/com.apple.CoreML/model.mlmodel",
+          "Data/com.apple.CoreML/weights/weight.bin"
+        ],
+        runtimes: ["python"]
+      },
+      openvino: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/openvino/camstack-ppocr-v5-det-mobile.xml"),
+        sizeMB: 3,
+        runtimes: ["python"]
+      }
+    }
+  },
+  {
+    id: "ppocr-v5-rec-mobile",
+    name: "PP-OCRv5 Recognition Mobile",
+    description: "PP-OCRv5 mobile text recognition \u2014 100+ languages, CTC decoding",
+    inputSize: { width: 320, height: 48 },
+    labels: OCR_TEXT_LABELS,
+    formats: {
+      onnx: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/onnx/camstack-ppocr-v5-rec-mobile.onnx"),
+        sizeMB: 8
+      },
+      coreml: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/coreml/camstack-ppocr-v5-rec-mobile.mlpackage"),
+        sizeMB: 4,
+        isDirectory: true,
+        files: [
+          "Manifest.json",
+          "Data/com.apple.CoreML/model.mlmodel",
+          "Data/com.apple.CoreML/weights/weight.bin"
+        ],
+        runtimes: ["python"]
+      },
+      openvino: {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/openvino/camstack-ppocr-v5-rec-mobile.xml"),
+        sizeMB: 4,
+        runtimes: ["python"]
+      }
+    },
+    extraFiles: [
+      {
+        url: hfModelUrl(HF_REPO, "generalOcr/ppocr-v5/camstack-ppocr-v5-keys.txt"),
+        filename: "camstack-ppocr-v5-keys.txt",
+        sizeMB: 0.1
+      }
+    ]
+  }
+];
 export {
   ANIMAL_TYPE_MODELS,
   AUDIO_CLASSIFICATION_MODELS,
@@ -76,12 +245,12 @@ export {
   BIRD_SPECIES_MODELS,
   BirdGlobalClassifierAddon,
   BirdNABirdsClassifierAddon,
-  CameraNativeDetectionAddon,
   FACE_DETECTION_MODELS,
   FACE_RECOGNITION_MODELS,
   FaceDetectionAddon,
   FaceRecognitionAddon,
-  MotionDetectionAddon,
+  GENERAL_OCR_MODELS,
+  MLPACKAGE_FILES,
   NodeInferenceEngine,
   OBJECT_DETECTION_MODELS,
   ObjectDetectionAddon,
@@ -91,6 +260,8 @@ export {
   PlateRecognitionAddon,
   PythonInferenceEngine,
   SEGMENTATION_MODELS,
+  SEGMENTATION_REFINER_MODELS,
+  VEHICLE_TYPE_MODELS,
   cosineSimilarity,
   cropRegion,
   ctcDecode,

package/dist/index.mjs.map

@@ -1 +1 @@
-{"version":3,"sources":[],"sourcesContent":[],"mappings":"","names":[]}
+{"version":3,"sources":["../src/catalogs/general-ocr-models.ts"],"sourcesContent":["import type { ModelCatalogEntry, LabelDefinition } from '@camstack/types'\nimport { hfModelUrl } from '@camstack/types'\n\nconst HF_REPO = 'camstack/camstack-models'\n\nconst OCR_TEXT_LABELS: readonly LabelDefinition[] = [\n  { id: 'text', name: 'Scene Text' },\n] as const\n\n/**\n * General-purpose OCR models for scene text recognition in camera feeds.\n * These complement the plate-specific PaddleOCR models for broader text detection\n * (signs, labels, addresses, etc.).\n */\nexport const GENERAL_OCR_MODELS: readonly ModelCatalogEntry[] = [\n  // ── OnnxTR / docTR — lightweight general scene text recognition ──\n  {\n    id: 'doctr-det-db-mobilenet',\n    name: 'docTR Detection MobileNet',\n    description: 'docTR DBNet MobileNet V3 — lightweight text region detection for scene text',\n    inputSize: { width: 1024, height: 1024 },\n    labels: OCR_TEXT_LABELS,\n    formats: {\n      onnx: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/onnx/camstack-doctr-det-db-mobilenet-v3.onnx'),\n        sizeMB: 15,\n      },\n      coreml: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/coreml/camstack-doctr-det-db-mobilenet-v3.mlpackage'),\n        sizeMB: 8,\n        isDirectory: true,\n        files: [\n          'Manifest.json',\n          'Data/com.apple.CoreML/model.mlmodel',\n          'Data/com.apple.CoreML/weights/weight.bin',\n        ],\n        runtimes: ['python'],\n      },\n      openvino: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/openvino/camstack-doctr-det-db-mobilenet-v3.xml'),\n        sizeMB: 8,\n        runtimes: ['python'],\n      },\n    },\n  },\n  {\n    id: 'doctr-rec-crnn-mobilenet',\n    name: 'docTR Recognition CRNN MobileNet',\n    description: 'docTR CRNN MobileNet V3 — lightweight text recognition for detected regions',\n    inputSize: { width: 128, height: 32 },\n    labels: OCR_TEXT_LABELS,\n    formats: {\n      onnx: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/onnx/camstack-doctr-rec-crnn-mobilenet-v3.onnx'),\n        sizeMB: 5,\n      },\n      coreml: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/coreml/camstack-doctr-rec-crnn-mobilenet-v3.mlpackage'),\n        sizeMB: 3,\n        isDirectory: true,\n        files: [\n          'Manifest.json',\n          'Data/com.apple.CoreML/model.mlmodel',\n          'Data/com.apple.CoreML/weights/weight.bin',\n        ],\n        runtimes: ['python'],\n      },\n      openvino: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/openvino/camstack-doctr-rec-crnn-mobilenet-v3.xml'),\n        sizeMB: 3,\n        runtimes: ['python'],\n      },\n    },\n  },\n  {\n    id: 'doctr-rec-parseq',\n    name: 'docTR Recognition PARSeq',\n    description: 'docTR PARSeq — high-accuracy scene text recognition (top ICDAR scores)',\n    inputSize: { width: 128, height: 32 },\n    labels: OCR_TEXT_LABELS,\n    formats: {\n      onnx: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/onnx/camstack-doctr-rec-parseq.onnx'),\n        sizeMB: 25,\n      },\n      coreml: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/coreml/camstack-doctr-rec-parseq.mlpackage'),\n        sizeMB: 13,\n        isDirectory: true,\n        files: [\n          'Manifest.json',\n          'Data/com.apple.CoreML/model.mlmodel',\n          'Data/com.apple.CoreML/weights/weight.bin',\n        ],\n        runtimes: ['python'],\n      },\n      openvino: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/doctr/openvino/camstack-doctr-rec-parseq.xml'),\n        sizeMB: 13,\n        runtimes: ['python'],\n      },\n    },\n  },\n\n  // ── PaddleOCR PP-OCRv5 Mobile — general-purpose text detection + recognition ──\n  {\n    id: 'ppocr-v5-det-mobile',\n    name: 'PP-OCRv5 Detection Mobile',\n    description: 'PP-OCRv5 mobile text detection — optimized for edge, 100+ languages',\n    inputSize: { width: 640, height: 640 },\n    labels: OCR_TEXT_LABELS,\n    formats: {\n      onnx: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/onnx/camstack-ppocr-v5-det-mobile.onnx'),\n        sizeMB: 6,\n      },\n      coreml: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/coreml/camstack-ppocr-v5-det-mobile.mlpackage'),\n        sizeMB: 3,\n        isDirectory: true,\n        files: [\n          'Manifest.json',\n          'Data/com.apple.CoreML/model.mlmodel',\n          'Data/com.apple.CoreML/weights/weight.bin',\n        ],\n        runtimes: ['python'],\n      },\n      openvino: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/openvino/camstack-ppocr-v5-det-mobile.xml'),\n        sizeMB: 3,\n        runtimes: ['python'],\n      },\n    },\n  },\n  {\n    id: 'ppocr-v5-rec-mobile',\n    name: 'PP-OCRv5 Recognition Mobile',\n    description: 'PP-OCRv5 mobile text recognition — 100+ languages, CTC decoding',\n    inputSize: { width: 320, height: 48 },\n    labels: OCR_TEXT_LABELS,\n    formats: {\n      onnx: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/onnx/camstack-ppocr-v5-rec-mobile.onnx'),\n        sizeMB: 8,\n      },\n      coreml: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/coreml/camstack-ppocr-v5-rec-mobile.mlpackage'),\n        sizeMB: 4,\n        isDirectory: true,\n        files: [\n          'Manifest.json',\n          'Data/com.apple.CoreML/model.mlmodel',\n          'Data/com.apple.CoreML/weights/weight.bin',\n        ],\n        runtimes: ['python'],\n      },\n      openvino: {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/openvino/camstack-ppocr-v5-rec-mobile.xml'),\n        sizeMB: 4,\n        runtimes: ['python'],\n      },\n    },\n    extraFiles: [\n      {\n        url: hfModelUrl(HF_REPO, 'generalOcr/ppocr-v5/camstack-ppocr-v5-keys.txt'),\n        filename: 'camstack-ppocr-v5-keys.txt',\n        sizeMB: 0.1,\n      },\n    ],\n  },\n] as const\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AACA,SAAS,kBAAkB;AAE3B,IAAM,UAAU;AAEhB,IAAM,kBAA8C;AAAA,EAClD,EAAE,IAAI,QAAQ,MAAM,aAAa;AACnC;AAOO,IAAM,qBAAmD;AAAA;AAAA,EAE9D;AAAA,IACE,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,aAAa;AAAA,IACb,WAAW,EAAE,OAAO,MAAM,QAAQ,KAAK;AAAA,IACvC,QAAQ;AAAA,IACR,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,KAAK,WAAW,SAAS,+DAA+D;AAAA,QACxF,QAAQ;AAAA,MACV;AAAA,MACA,QAAQ;AAAA,QACN,KAAK,WAAW,SAAS,sEAAsE;AAAA,QAC/F,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,OAAO;AAAA,UACL;AAAA,UACA;AAAA,UACA;AAAA,QACF;AAAA,QACA,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,MACA,UAAU;AAAA,QACR,KAAK,WAAW,SAAS,kEAAkE;AAAA,QAC3F,QAAQ;AAAA,QACR,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,IACF;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,aAAa;AAAA,IACb,WAAW,EAAE,OAAO,KAAK,QAAQ,GAAG;AAAA,IACpC,QAAQ;AAAA,IACR,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,KAAK,WAAW,SAAS,iEAAiE;AAAA,QAC1F,QAAQ;AAAA,MACV;AAAA,MACA,QAAQ;AAAA,QACN,KAAK,WAAW,SAAS,wEAAwE;AAAA,QACjG,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,OAAO;AAAA,UACL;AAAA,UACA;AAAA,UACA;AAAA,QACF;AAAA,QACA,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,MACA,UAAU;AAAA,QACR,KAAK,WAAW,SAAS,oEAAoE;AAAA,QAC7F,QAAQ;AAAA,QACR,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,IACF;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,aAAa;AAAA,IACb,WAAW,EAAE,OAAO,KAAK,QAAQ,GAAG;AAAA,IACpC,QAAQ;AAAA,IACR,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,KAAK,WAAW,SAAS,sDAAsD;AAAA,QAC/E,QAAQ;AAAA,MACV;AAAA,MACA,QAAQ;AAAA,QACN,KAAK,WAAW,SAAS,6DAA6D;AAAA,QACtF,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,OAAO;AAAA,UACL;AAAA,UACA;AAAA,UACA;AAAA,QACF;AAAA,QACA,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,MACA,UAAU;AAAA,QACR,KAAK,WAAW,SAAS,yDAAyD;AAAA,QAClF,QAAQ;AAAA,QACR,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,IACF;AAAA,EACF;AAAA;AAAA,EAGA;AAAA,IACE,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,aAAa;AAAA,IACb,WAAW,EAAE,OAAO,KAAK,QAAQ,IAAI;AAAA,IACrC,QAAQ;AAAA,IACR,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,KAAK,WAAW,SAAS,4DAA4D;AAAA,QACrF,QAAQ;AAAA,MACV;AAAA,MACA,QAAQ;AAAA,QACN,KAAK,WAAW,SAAS,mEAAmE;AAAA,QAC5F,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,OAAO;AAAA,UACL;AAAA,UACA;AAAA,UACA;AAAA,QACF;AAAA,QACA,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,MACA,UAAU;AAAA,QACR,KAAK,WAAW,SAAS,+DAA+D;AAAA,QACxF,QAAQ;AAAA,QACR,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,IACF;AAAA,EACF;AAAA,EACA;AAAA,IACE,IAAI;AAAA,IACJ,MAAM;AAAA,IACN,aAAa;AAAA,IACb,WAAW,EAAE,OAAO,KAAK,QAAQ,GAAG;AAAA,IACpC,QAAQ;AAAA,IACR,SAAS;AAAA,MACP,MAAM;AAAA,QACJ,KAAK,WAAW,SAAS,4DAA4D;AAAA,QACrF,QAAQ;AAAA,MACV;AAAA,MACA,QAAQ;AAAA,QACN,KAAK,WAAW,SAAS,mEAAmE;AAAA,QAC5F,QAAQ;AAAA,QACR,aAAa;AAAA,QACb,OAAO;AAAA,UACL;AAAA,UACA;AAAA,UACA;AAAA,QACF;AAAA,QACA,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,MACA,UAAU;AAAA,QACR,KAAK,WAAW,SAAS,+DAA+D;AAAA,QACxF,QAAQ;AAAA,QACR,UAAU,CAAC,QAAQ;AAAA,MACrB;AAAA,IACF;AAAA,IACA,YAAY;AAAA,MACV;AAAA,QACE,KAAK,WAAW,SAAS,gDAAgD;AAAA,QACzE,UAAU;AAAA,QACV,QAAQ;AAAA,MACV;AAAA,IACF;AAAA,EACF;AACF;","names":[]}

package/package.json

@@ -1,8 +1,17 @@
 {
   "name": "@camstack/addon-vision",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Builtin detection addons for CamStack — object detection, face, plate, audio, motion",
-  "keywords": ["camstack", "addon", "camstack-addon", "vision", "object-detection", "face-detection", "motion-detection", "camera"],
+  "keywords": [
+    "camstack",
+    "addon",
+    "camstack-addon",
+    "vision",
+    "object-detection",
+    "face-detection",
+    "motion-detection",
+    "camera"
+  ],
   "license": "MIT",
   "repository": {
     "type": "git",
@@ -12,25 +21,87 @@
   "module": "./dist/index.mjs",
   "types": "./dist/index.d.ts",
   "exports": {
-    ".": { "import": "./dist/index.mjs", "require": "./dist/index.js", "types": "./dist/index.d.ts" },
-    "./addons/*": { "import": "./dist/addons/*/index.mjs", "require": "./dist/addons/*/index.js", "types": "./dist/addons/*/index.d.ts" }
+    ".": {
+      "import": "./dist/index.mjs",
+      "require": "./dist/index.js",
+      "types": "./dist/index.d.ts"
+    },
+    "./addons/*": {
+      "import": "./dist/addons/*/index.mjs",
+      "require": "./dist/addons/*/index.js",
+      "types": "./dist/addons/*/index.d.ts"
+    },
+    "./package.json": "./package.json"
   },
   "camstack": {
+    "displayName": "CamStack Vision",
     "addons": [
-      { "id": "motion-detection", "entry": "./dist/addons/motion-detection/index.js", "slot": "detector" },
-      { "id": "object-detection", "entry": "./dist/addons/object-detection/index.js", "slot": "detector" },
-      { "id": "face-detection", "entry": "./dist/addons/face-detection/index.js", "slot": "cropper" },
-      { "id": "face-recognition", "entry": "./dist/addons/face-recognition/index.js", "slot": "classifier" },
-      { "id": "plate-detection", "entry": "./dist/addons/plate-detection/index.js", "slot": "cropper" },
-      { "id": "plate-recognition", "entry": "./dist/addons/plate-recognition/index.js", "slot": "classifier" },
-      { "id": "audio-classification", "entry": "./dist/addons/audio-classification/index.js", "slot": "classifier" },
-      { "id": "camera-native-detection", "entry": "./dist/addons/camera-native-detection/index.js", "slot": "detector" },
-      { "id": "bird-global-classifier", "entry": "./dist/addons/bird-global-classifier/index.js", "slot": "classifier" },
-      { "id": "bird-nabirds-classifier", "entry": "./dist/addons/bird-nabirds-classifier/index.js", "slot": "classifier" },
-      { "id": "animal-classifier", "entry": "./dist/addons/animal-classifier/index.js", "slot": "classifier" }
+      {
+        "id": "motion-detection",
+        "entry": "./dist/addons/motion-detection/index.js",
+        "slot": "detector"
+      },
+      {
+        "id": "object-detection",
+        "entry": "./dist/addons/object-detection/index.js",
+        "slot": "detector"
+      },
+      {
+        "id": "face-detection",
+        "entry": "./dist/addons/face-detection/index.js",
+        "slot": "cropper"
+      },
+      {
+        "id": "face-recognition",
+        "entry": "./dist/addons/face-recognition/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "plate-detection",
+        "entry": "./dist/addons/plate-detection/index.js",
+        "slot": "cropper"
+      },
+      {
+        "id": "plate-recognition",
+        "entry": "./dist/addons/plate-recognition/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "audio-classification",
+        "entry": "./dist/addons/audio-classification/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "bird-global-classifier",
+        "entry": "./dist/addons/bird-global-classifier/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "bird-nabirds-classifier",
+        "entry": "./dist/addons/bird-nabirds-classifier/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "animal-classifier",
+        "entry": "./dist/addons/animal-classifier/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "vehicle-classifier",
+        "entry": "./dist/addons/vehicle-classifier/index.js",
+        "slot": "classifier"
+      },
+      {
+        "id": "segmentation-refiner",
+        "entry": "./dist/addons/segmentation-refiner/index.js",
+        "slot": "refiner"
+      }
     ]
   },
-  "files": ["dist", "python"],
+  "files": [
+    "dist",
+    "python"
+  ],
   "scripts": {
     "build": "tsup",
     "dev": "tsup --watch",
@@ -38,8 +109,13 @@
     "test": "vitest run",
     "test:watch": "vitest"
   },
-  "peerDependencies": { "@camstack/types": "^0.1.0" },
-  "dependencies": { "onnxruntime-node": "^1.24.3", "sharp": "^0.34.0" },
+  "peerDependencies": {
+    "@camstack/types": "^0.1.0"
+  },
+  "dependencies": {
+    "onnxruntime-node": "^1.24.3",
+    "sharp": "^0.34.0"
+  },
   "devDependencies": {
     "@camstack/types": "*",
     "tsup": "^8.0.0",

package/python/coreml_inference.py

@@ -110,15 +110,23 @@ def compute_iou(x1a: float, y1a: float, x2a: float, y2a: float,
 def parse_yolo_output(output: np.ndarray, conf_threshold: float,
                       img_w: int, img_h: int, input_size: int,
                       scale: float, pad: tuple[int, int]) -> list[dict]:
-    """Parse YOLO output tensor [1, 84, 8400] into normalised detections."""
-    # Squeeze batch dim and transpose to [num_boxes, 84]
+    """Parse YOLO output tensor [1, C, 8400] into normalised detections.
+
+    Supports both detection (C=84) and segmentation (C=116) outputs.
+    For seg models the extra 32 mask coefficients are ignored (detections only).
+    """
+    # Squeeze batch dim
     if output.ndim == 3 and output.shape[0] == 1:
         output = output[0]
-    if output.shape[0] == 84:
-        output = output.T  # [8400, 84]
+
+    # Transpose if channels-first: [C, N] -> [N, C]
+    num_channels = output.shape[0]
+    if num_channels in (84, 116, 144):  # det=84, seg=116, etc.
+        output = output.T  # [8400, C]
 
     cx, cy, w, h = output[:, 0], output[:, 1], output[:, 2], output[:, 3]
-    class_scores = output[:, 4:]  # [num_boxes, 80]
+    # Take only the 80 class scores (skip mask coefficients if seg model)
+    class_scores = output[:, 4:84]  # [num_boxes, 80] — works for both det and seg
 
     class_ids = np.argmax(class_scores, axis=1)
     scores = class_scores[np.arange(len(class_ids)), class_ids]
@@ -179,8 +187,12 @@ def parse_yolo_output(output: np.ndarray, conf_threshold: float,
 # ---------------------------------------------------------------------------
 
 def _parse_nms_output(coords: np.ndarray, confs: np.ndarray,
-                      conf_threshold: float, img_w: int, img_h: int) -> list[dict]:
-    """Parse CoreML NMS output: (N, 4) coords [cx, cy, w, h] normalized + (N, 80) scores."""
+                      conf_threshold: float, img_w: int, img_h: int,
+                      input_size: int, scale: float, pad: tuple[int, int]) -> list[dict]:
+    """Parse CoreML NMS output: (N, 4) coords [cx, cy, w, h] normalized 0-1 relative to model input.
+
+    Coordinates must be un-letterboxed to get correct positions in the original image.
+    """
     if coords.shape[0] == 0:
         return []
 
@@ -195,12 +207,31 @@ def _parse_nms_output(coords: np.ndarray, confs: np.ndarray,
     if len(scores) == 0:
         return []
 
-    # CoreML coords are [cx, cy, w, h] already normalized 0-1
-    cx, cy, w, h = coords[:, 0], coords[:, 1], coords[:, 2], coords[:, 3]
-    x1 = np.clip(cx - w / 2, 0, 1)
-    y1 = np.clip(cy - h / 2, 0, 1)
-    x2 = np.clip(cx + w / 2, 0, 1)
-    y2 = np.clip(cy + h / 2, 0, 1)
+    # CoreML NMS coords are [cx, cy, w, h] normalized 0-1 relative to model input (640x640)
+    # Convert to pixel coords in the letterboxed frame first
+    cx = coords[:, 0] * input_size
+    cy = coords[:, 1] * input_size
+    bw = coords[:, 2] * input_size
+    bh = coords[:, 3] * input_size
+
+    # Corner format
+    x1 = cx - bw / 2
+    y1 = cy - bh / 2
+    x2 = cx + bw / 2
+    y2 = cy + bh / 2
+
+    # Undo letterbox padding and scale → original image pixel coords
+    pad_x, pad_y = pad
+    x1 = (x1 - pad_x) / scale
+    y1 = (y1 - pad_y) / scale
+    x2 = (x2 - pad_x) / scale
+    y2 = (y2 - pad_y) / scale
+
+    # Normalize to 0-1 relative to original image
+    x1 = np.clip(x1 / img_w, 0, 1)
+    y1 = np.clip(y1 / img_h, 0, 1)
+    x2 = np.clip(x2 / img_w, 0, 1)
+    y2 = np.clip(y2 / img_h, 0, 1)
 
     detections = []
     for i in range(len(scores)):
@@ -293,13 +324,25 @@ def main() -> None:
 
             if has_builtin_nms:
                 # Model has built-in NMS: outputs are (N, 4) boxes + (N, 80) scores
-                coords = np.array(predictions['coordinates'])   # (N, 4) — cx, cy, w, h normalized
+                coords = np.array(predictions['coordinates'])   # (N, 4) — cx, cy, w, h normalized to model input
                 confs = np.array(predictions['confidence'])     # (N, 80)
-                detections = _parse_nms_output(coords, confs, args.confidence, orig_w, orig_h)
+                detections = _parse_nms_output(coords, confs, args.confidence, orig_w, orig_h,
+                                               args.input_size, scale, pad)
             else:
-                # Raw YOLO output: [1, 84, 8400]
-                output_key = list(predictions.keys())[0]
-                output = np.array(predictions[output_key])
+                # Raw YOLO output: [1, C, N] where C=84 (det) or C=116 (seg)
+                # For seg models with multiple outputs, pick the detection tensor
+                # (the one with shape [1, 84|116, 8400], not the mask protos [1, 32, 160, 160])
+                output_keys = list(predictions.keys())
+                output = None
+                for key in output_keys:
+                    arr = np.array(predictions[key])
+                    # Detection tensor has a dimension of 84 or 116 (not 32, not 160)
+                    if arr.ndim >= 2 and any(d in (84, 116, 144) for d in arr.shape):
+                        output = arr
+                        break
+                if output is None:
+                    output = np.array(predictions[output_keys[0]])
+
                 detections = parse_yolo_output(
                     output, args.confidence, orig_w, orig_h,
                     args.input_size, scale, pad,

package/python/openvino_inference.py

@@ -110,14 +110,22 @@ def compute_iou(x1a: float, y1a: float, x2a: float, y2a: float,
 def parse_yolo_output(output: np.ndarray, conf_threshold: float,
                       img_w: int, img_h: int, input_size: int,
                       scale: float, pad: tuple[int, int]) -> list[dict]:
-    """Parse YOLO output tensor [1, 84, 8400] into normalised detections."""
+    """Parse YOLO output tensor [1, C, 8400] into normalised detections.
+
+    Supports both detection (C=84) and segmentation (C=116) outputs.
+    For seg models the extra 32 mask coefficients are ignored (detections only).
+    """
     if output.ndim == 3 and output.shape[0] == 1:
         output = output[0]
-    if output.shape[0] == 84:
-        output = output.T  # [8400, 84]
+
+    # Transpose if channels-first: [C, N] -> [N, C]
+    num_channels = output.shape[0]
+    if num_channels in (84, 116, 144):  # det=84, seg=116, etc.
+        output = output.T  # [8400, C]
 
     cx, cy, w, h = output[:, 0], output[:, 1], output[:, 2], output[:, 3]
-    class_scores = output[:, 4:]
+    # Take only the 80 class scores (skip mask coefficients if seg model)
+    class_scores = output[:, 4:84]  # [num_boxes, 80] — works for both det and seg
 
     class_ids = np.argmax(class_scores, axis=1)
     scores = class_scores[np.arange(len(class_ids)), class_ids]

package/python/pytorch_inference.py

@@ -114,14 +114,22 @@ def compute_iou(x1a: float, y1a: float, x2a: float, y2a: float,
 def parse_yolo_output(output: np.ndarray, conf_threshold: float,
                       img_w: int, img_h: int, input_size: int,
                       scale: float, pad: tuple[int, int]) -> list[dict]:
-    """Parse YOLO output tensor [1, 84, 8400] into normalised detections."""
+    """Parse YOLO output tensor [1, C, 8400] into normalised detections.
+
+    Supports both detection (C=84) and segmentation (C=116) outputs.
+    For seg models the extra 32 mask coefficients are ignored (detections only).
+    """
     if output.ndim == 3 and output.shape[0] == 1:
         output = output[0]
-    if output.shape[0] == 84:
-        output = output.T  # [8400, 84]
+
+    # Transpose if channels-first: [C, N] -> [N, C]
+    num_channels = output.shape[0]
+    if num_channels in (84, 116, 144):  # det=84, seg=116, etc.
+        output = output.T  # [8400, C]
 
     cx, cy, w, h = output[:, 0], output[:, 1], output[:, 2], output[:, 3]
-    class_scores = output[:, 4:]
+    # Take only the 80 class scores (skip mask coefficients if seg model)
+    class_scores = output[:, 4:84]  # [num_boxes, 80] — works for both det and seg
 
     class_ids = np.argmax(class_scores, axis=1)
     scores = class_scores[np.arange(len(class_ids)), class_ids]

package/dist/addons/camera-native-detection/index.d.mts

@@ -1,32 +0,0 @@
-import { IDetectorProvider, IDetectionAddon, AddonManifest, AddonContext, FrameInput, DetectorOutput, ConfigUISchema, ClassMapDefinition, ModelCatalogEntry, DetectionModel, LabelDefinition, ProbeResult } from '@camstack/types';
-
-/**
- * CameraNativeDetectionAddon
- *
- * A stub detector that wraps native camera event subscriptions (e.g. Frigate / Scrypted
- * webhooks). The detect() method always returns empty — real detections are delivered
- * asynchronously via external event subscriptions which should populate the pipeline
- * from outside this addon's detect() call.
- *
- * This addon exists so the pipeline can declare a 'detector' slot backed by camera events
- * without requiring any inference model.
- */
-declare class CameraNativeDetectionAddon implements IDetectorProvider, IDetectionAddon {
-    readonly id = "camera-native-detection";
-    readonly slot: "detector";
-    readonly inputClasses: readonly string[] | null;
-    readonly outputClasses: readonly ["person", "vehicle", "motion", "face"];
-    readonly slotPriority = 5;
-    readonly manifest: AddonManifest;
-    initialize(_ctx: AddonContext): Promise<void>;
-    detect(_frame: FrameInput): Promise<DetectorOutput>;
-    shutdown(): Promise<void>;
-    getConfigSchema(): ConfigUISchema;
-    getClassMap(): ClassMapDefinition;
-    getModelCatalog(): ModelCatalogEntry[];
-    getAvailableModels(): DetectionModel[];
-    getActiveLabels(): readonly LabelDefinition[];
-    probe(): Promise<ProbeResult>;
-}
-
-export { CameraNativeDetectionAddon as default };