@camstack/addon-pipeline 1.1.53 → 1.1.55
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/audio-analyzer/index.js +1 -1
- package/dist/audio-analyzer/index.mjs +1 -1
- package/dist/detection-pipeline/index.js +289 -1469
- package/dist/detection-pipeline/index.mjs +269 -1449
- package/dist/{dist-DWb62H5U.js → dist-DI57FC8K.js} +167 -8
- package/dist/{dist-BalxNwOt.mjs → dist-RWGGPwVx.mjs} +167 -8
- package/dist/motion-wasm/index.js +1 -1
- package/dist/motion-wasm/index.mjs +1 -1
- package/dist/pipeline-runner/index.js +772 -21
- package/dist/pipeline-runner/index.mjs +772 -22
- package/dist/recorder/index.js +1 -1
- package/dist/recorder/index.mjs +1 -1
- package/dist/{remote-source-plane-CZpzIVro.js → remote-source-plane-CHgvzzA6.js} +1 -1
- package/dist/{remote-source-plane-BCJW5CvF.mjs → remote-source-plane-DU0aRSPv.mjs} +1 -1
- package/dist/session-decode/decode-worker-child.js +538 -35
- package/dist/session-decode/decode-worker-child.mjs +538 -35
- package/dist/step-definitions-CNBFKjZe.js +1514 -0
- package/dist/step-definitions-CP9kVSml.mjs +1479 -0
- package/dist/stream-broker/_stub.js +2 -2
- package/dist/stream-broker/{_virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-DIN3CcRP.mjs → _virtual_mf-localSharedImportMap___mfe_internal__addon_stream_broker_widgets-mDCPzmT3.mjs} +3 -3
- package/dist/stream-broker/_virtual_mf___mfe_internal__addon_stream_broker_widgets__loadShare___mf_0_camstack_mf_1_types__loadShare__.js-B3gTdHEh.mjs +26 -0
- package/dist/stream-broker/{_virtual_mf___mfe_internal__addon_stream_broker_widgets__loadShare___mf_0_camstack_mf_1_ui_mf_2_library__loadShare__.js-C9fwKMfg.mjs → _virtual_mf___mfe_internal__addon_stream_broker_widgets__loadShare___mf_0_camstack_mf_1_ui_mf_2_library__loadShare__.js-Dn_pxzP-.mjs} +1 -1
- package/dist/stream-broker/{hostInit-CGUEq--X.mjs → hostInit-GM_CI22k.mjs} +3 -3
- package/dist/stream-broker/index.js +2 -2
- package/dist/stream-broker/index.mjs +2 -2
- package/dist/stream-broker/remoteEntry.js +1 -1
- package/dist/{worker-protocol-pk7qdYXt.mjs → worker-protocol-CyVJTZEO.mjs} +7 -0
- package/dist/{worker-protocol-BCfO8gUF.js → worker-protocol-PP4jKHHJ.js} +7 -0
- package/embed-dist/assets/{MaskShapeCanvas-DI4BY7W2-CqOe8eYa.js → MaskShapeCanvas-DI4BY7W2-BChW0ntM.js} +1 -1
- package/embed-dist/assets/{MotionZonesSettings-NcxxQN8r-CZyLeUnd.js → MotionZonesSettings-NcxxQN8r-DDzqEbSe.js} +1 -1
- package/embed-dist/assets/{PrivacyMaskSettings-APgPLF7p-Cn0hGZnu.js → PrivacyMaskSettings-APgPLF7p-B879GXaf.js} +1 -1
- package/embed-dist/assets/{index-DRl4XYjA.js → index-CNjQ5rAE.js} +10 -10
- package/embed-dist/index.html +1 -1
- package/package.json +1 -1
- package/dist/stream-broker/_virtual_mf___mfe_internal__addon_stream_broker_widgets__loadShare___mf_0_camstack_mf_1_types__loadShare__.js-o4tu_xuc.mjs +0 -26
|
@@ -0,0 +1,1479 @@
|
|
|
1
|
+
import { S as hfModelUrl, a as COCO_TO_MACRO, i as COCO_80_LABELS, r as AUDIO_MACRO_LABELS } from "./dist-RWGGPwVx.mjs";
|
|
2
|
+
//#region src/detection-pipeline/registry/model-catalogs.ts
|
|
3
|
+
var HF_REPO = "camstack/camstack-models";
|
|
4
|
+
var HF_SCRYPTED = "scrypted/plugin-models";
|
|
5
|
+
var hf = (path) => hfModelUrl(HF_REPO, path);
|
|
6
|
+
var hfScrypted = (path) => hfModelUrl(HF_SCRYPTED, path);
|
|
7
|
+
/**
|
|
8
|
+
* Build an OpenVINO format entry (always python runtime).
|
|
9
|
+
*
|
|
10
|
+
* OpenVINO IR is a two-file bundle: a `.xml` topology + a sibling `.bin`
|
|
11
|
+
* weights file with the same basename. We declare the `.bin` in `files` so
|
|
12
|
+
* the (format-agnostic) downloader fetches it alongside the `.xml` — without
|
|
13
|
+
* the weights, OpenVINO compile fails with "Empty weights data in bin file".
|
|
14
|
+
* A plain `.onnx` run through the OpenVINO runtime (e.g. yamnet) has no
|
|
15
|
+
* sibling, so none is added.
|
|
16
|
+
*/
|
|
17
|
+
var ovFormat = (url, sizeMB) => {
|
|
18
|
+
const base = url.split("/").pop() ?? "";
|
|
19
|
+
const files = base.endsWith(".xml") ? [base.replace(/\.xml$/, ".bin")] : void 0;
|
|
20
|
+
return {
|
|
21
|
+
url,
|
|
22
|
+
sizeMB,
|
|
23
|
+
runtimes: ["python"],
|
|
24
|
+
...files ? { files } : {}
|
|
25
|
+
};
|
|
26
|
+
};
|
|
27
|
+
/**
|
|
28
|
+
* Build a precision-variant catalog entry (OpenVINO-only) derived from a base
|
|
29
|
+
* detection model. fp16 halves the weights (Intel iGPU/NPU sweet spot); int8 is
|
|
30
|
+
* NNCF post-training-quantized (~4× smaller, fastest on CPU/iGPU at a small
|
|
31
|
+
* accuracy cost). The IRs live next to the base `.xml` on HF as
|
|
32
|
+
* `camstack-<id>-<precision>.xml`. Lets an operator scale the model to the node
|
|
33
|
+
* (e.g. yolo26x-int8 on a 265K, yolo26n-int8 on an N100).
|
|
34
|
+
*/
|
|
35
|
+
var ovPrecisionVariant = (baseId, ovDir, baseName, precision, sizeMB, legacy = false) => ({
|
|
36
|
+
id: `${baseId}-${precision}`,
|
|
37
|
+
name: `${baseName} (${precision.toUpperCase()})`,
|
|
38
|
+
description: `${baseName} — OpenVINO ${precision.toUpperCase()} variant for Intel iGPU/NPU; scale by hardware`,
|
|
39
|
+
inputSize: {
|
|
40
|
+
width: 640,
|
|
41
|
+
height: 640
|
|
42
|
+
},
|
|
43
|
+
labels: [],
|
|
44
|
+
preprocessMode: "letterbox",
|
|
45
|
+
formats: { openvino: ovFormat(hf(`${ovDir}/camstack-${baseId}-${precision}.xml`), sizeMB) },
|
|
46
|
+
...legacy ? { legacy: true } : {},
|
|
47
|
+
...precision === "int8" ? { group: {
|
|
48
|
+
family: baseId.slice(0, -1),
|
|
49
|
+
tier: baseId.slice(-1),
|
|
50
|
+
precision: "int8"
|
|
51
|
+
} } : {}
|
|
52
|
+
});
|
|
53
|
+
var MLPACKAGE_FILES = [
|
|
54
|
+
"Manifest.json",
|
|
55
|
+
"Data/com.apple.CoreML/model.mlmodel",
|
|
56
|
+
"Data/com.apple.CoreML/weights/weight.bin"
|
|
57
|
+
];
|
|
58
|
+
var YOLO26_TIER_NAME = {
|
|
59
|
+
n: "Nano",
|
|
60
|
+
s: "Small",
|
|
61
|
+
m: "Medium",
|
|
62
|
+
l: "Large"
|
|
63
|
+
};
|
|
64
|
+
/**
|
|
65
|
+
* Reduced-input YOLO26 variants (2026-07-11). A lower square input (320 / 256)
|
|
66
|
+
* is the biggest, cheapest latency lever measured across every runtime — Mac
|
|
67
|
+
* ANE, Intel Ultra NPU, and especially the N100 — at a small-object accuracy
|
|
68
|
+
* cost. Each resolution ships a base (fp32) entry (onnx + coreml + openvino)
|
|
69
|
+
* grouped under `{family, tier, resolution}`, plus an OpenVINO INT8 entry (best
|
|
70
|
+
* on Intel/N100) grouped with `precision:'int8'`. The grouped selector surfaces
|
|
71
|
+
* `resolution` as a variant axis alongside precision.
|
|
72
|
+
*/
|
|
73
|
+
var yolo26Reduced = (tier, res, sizes) => {
|
|
74
|
+
const name = YOLO26_TIER_NAME[tier] ?? tier.toUpperCase();
|
|
75
|
+
return [{
|
|
76
|
+
id: `yolo26${tier}-${res}`,
|
|
77
|
+
group: {
|
|
78
|
+
family: "yolo26",
|
|
79
|
+
tier,
|
|
80
|
+
resolution: res
|
|
81
|
+
},
|
|
82
|
+
name: `YOLO26 ${name} @${res}`,
|
|
83
|
+
description: `YOLO26 ${name}, reduced ${res}×${res} input — faster on ANE / hub NPU / N100 at a small-object accuracy cost`,
|
|
84
|
+
inputSize: {
|
|
85
|
+
width: res,
|
|
86
|
+
height: res
|
|
87
|
+
},
|
|
88
|
+
labels: [],
|
|
89
|
+
preprocessMode: "letterbox",
|
|
90
|
+
formats: {
|
|
91
|
+
onnx: {
|
|
92
|
+
url: hf(`objectDetection/yolo26/onnx/camstack-yolo26${tier}-${res}.onnx`),
|
|
93
|
+
sizeMB: sizes.onnx
|
|
94
|
+
},
|
|
95
|
+
coreml: {
|
|
96
|
+
url: hf(`objectDetection/yolo26/coreml/camstack-yolo26${tier}-${res}.mlpackage`),
|
|
97
|
+
sizeMB: sizes.coreml,
|
|
98
|
+
isDirectory: true,
|
|
99
|
+
files: [...MLPACKAGE_FILES],
|
|
100
|
+
runtimes: ["python"]
|
|
101
|
+
},
|
|
102
|
+
openvino: ovFormat(hf(`objectDetection/yolo26/openvino/camstack-yolo26${tier}-${res}.xml`), sizes.ovFp32)
|
|
103
|
+
}
|
|
104
|
+
}, {
|
|
105
|
+
id: `yolo26${tier}-${res}-int8`,
|
|
106
|
+
group: {
|
|
107
|
+
family: "yolo26",
|
|
108
|
+
tier,
|
|
109
|
+
precision: "int8",
|
|
110
|
+
resolution: res
|
|
111
|
+
},
|
|
112
|
+
name: `YOLO26 ${name} @${res} (INT8)`,
|
|
113
|
+
description: `YOLO26 ${name} @${res} — OpenVINO INT8 for Intel iGPU/CPU; the N100's fastest config`,
|
|
114
|
+
inputSize: {
|
|
115
|
+
width: res,
|
|
116
|
+
height: res
|
|
117
|
+
},
|
|
118
|
+
labels: [],
|
|
119
|
+
preprocessMode: "letterbox",
|
|
120
|
+
formats: { openvino: ovFormat(hf(`objectDetection/yolo26/openvino/camstack-yolo26${tier}-${res}-int8.xml`), sizes.ovInt8) }
|
|
121
|
+
}];
|
|
122
|
+
};
|
|
123
|
+
var OBJECT_DETECTION_MODELS = [
|
|
124
|
+
{
|
|
125
|
+
id: "yolov9t",
|
|
126
|
+
legacy: true,
|
|
127
|
+
name: "YOLOv9 Tiny",
|
|
128
|
+
description: "YOLOv9 Tiny — ultra-lightweight next-gen detector",
|
|
129
|
+
inputSize: {
|
|
130
|
+
width: 640,
|
|
131
|
+
height: 640
|
|
132
|
+
},
|
|
133
|
+
labels: [],
|
|
134
|
+
preprocessMode: "letterbox",
|
|
135
|
+
formats: {
|
|
136
|
+
onnx: {
|
|
137
|
+
url: hf("objectDetection/yolov9/onnx/camstack-yolov9t.onnx"),
|
|
138
|
+
sizeMB: 8
|
|
139
|
+
},
|
|
140
|
+
coreml: {
|
|
141
|
+
url: hf("objectDetection/yolov9/coreml/camstack-yolov9t.mlpackage"),
|
|
142
|
+
sizeMB: 4,
|
|
143
|
+
isDirectory: true,
|
|
144
|
+
files: [...MLPACKAGE_FILES],
|
|
145
|
+
runtimes: ["python"]
|
|
146
|
+
},
|
|
147
|
+
openvino: ovFormat(hf("objectDetection/yolov9/openvino/camstack-yolov9t.xml"), 6)
|
|
148
|
+
}
|
|
149
|
+
},
|
|
150
|
+
{
|
|
151
|
+
id: "yolov9s",
|
|
152
|
+
legacy: true,
|
|
153
|
+
name: "YOLOv9 Small",
|
|
154
|
+
description: "YOLOv9 Small — improved efficiency over YOLOv8s",
|
|
155
|
+
inputSize: {
|
|
156
|
+
width: 640,
|
|
157
|
+
height: 640
|
|
158
|
+
},
|
|
159
|
+
labels: [],
|
|
160
|
+
preprocessMode: "letterbox",
|
|
161
|
+
formats: {
|
|
162
|
+
onnx: {
|
|
163
|
+
url: hf("objectDetection/yolov9/onnx/camstack-yolov9s.onnx"),
|
|
164
|
+
sizeMB: 28
|
|
165
|
+
},
|
|
166
|
+
coreml: {
|
|
167
|
+
url: hf("objectDetection/yolov9/coreml/camstack-yolov9s.mlpackage"),
|
|
168
|
+
sizeMB: 14,
|
|
169
|
+
isDirectory: true,
|
|
170
|
+
files: [...MLPACKAGE_FILES],
|
|
171
|
+
runtimes: ["python"]
|
|
172
|
+
},
|
|
173
|
+
openvino: ovFormat(hf("objectDetection/yolov9/openvino/camstack-yolov9s.xml"), 16)
|
|
174
|
+
}
|
|
175
|
+
},
|
|
176
|
+
{
|
|
177
|
+
id: "yolov9c",
|
|
178
|
+
legacy: true,
|
|
179
|
+
name: "YOLOv9 C",
|
|
180
|
+
description: "YOLOv9 C — high-accuracy compact model",
|
|
181
|
+
inputSize: {
|
|
182
|
+
width: 640,
|
|
183
|
+
height: 640
|
|
184
|
+
},
|
|
185
|
+
labels: [],
|
|
186
|
+
preprocessMode: "letterbox",
|
|
187
|
+
formats: {
|
|
188
|
+
onnx: {
|
|
189
|
+
url: hf("objectDetection/yolov9/onnx/camstack-yolov9c.onnx"),
|
|
190
|
+
sizeMB: 97
|
|
191
|
+
},
|
|
192
|
+
coreml: {
|
|
193
|
+
url: hf("objectDetection/yolov9/coreml/camstack-yolov9c.mlpackage"),
|
|
194
|
+
sizeMB: 48,
|
|
195
|
+
isDirectory: true,
|
|
196
|
+
files: [...MLPACKAGE_FILES],
|
|
197
|
+
runtimes: ["python"]
|
|
198
|
+
},
|
|
199
|
+
openvino: ovFormat(hf("objectDetection/yolov9/openvino/camstack-yolov9c.xml"), 49)
|
|
200
|
+
}
|
|
201
|
+
},
|
|
202
|
+
{
|
|
203
|
+
id: "yolo26n",
|
|
204
|
+
group: {
|
|
205
|
+
family: "yolo26",
|
|
206
|
+
tier: "n"
|
|
207
|
+
},
|
|
208
|
+
name: "YOLO26 Nano",
|
|
209
|
+
description: "YOLO26 Nano — latest generation ultra-lightweight detector with attention mechanism",
|
|
210
|
+
inputSize: {
|
|
211
|
+
width: 640,
|
|
212
|
+
height: 640
|
|
213
|
+
},
|
|
214
|
+
labels: [],
|
|
215
|
+
preprocessMode: "letterbox",
|
|
216
|
+
formats: {
|
|
217
|
+
onnx: {
|
|
218
|
+
url: hf("objectDetection/yolo26/onnx/camstack-yolo26n.onnx"),
|
|
219
|
+
sizeMB: 10
|
|
220
|
+
},
|
|
221
|
+
coreml: {
|
|
222
|
+
url: hf("objectDetection/yolo26/coreml/camstack-yolo26n.mlpackage"),
|
|
223
|
+
sizeMB: 5,
|
|
224
|
+
isDirectory: true,
|
|
225
|
+
files: [...MLPACKAGE_FILES],
|
|
226
|
+
runtimes: ["python"]
|
|
227
|
+
},
|
|
228
|
+
openvino: ovFormat(hf("objectDetection/yolo26/openvino/camstack-yolo26n.xml"), 9)
|
|
229
|
+
}
|
|
230
|
+
},
|
|
231
|
+
{
|
|
232
|
+
id: "yolo26s",
|
|
233
|
+
group: {
|
|
234
|
+
family: "yolo26",
|
|
235
|
+
tier: "s"
|
|
236
|
+
},
|
|
237
|
+
name: "YOLO26 Small",
|
|
238
|
+
description: "YOLO26 Small — balanced speed/accuracy with attention-based architecture",
|
|
239
|
+
inputSize: {
|
|
240
|
+
width: 640,
|
|
241
|
+
height: 640
|
|
242
|
+
},
|
|
243
|
+
labels: [],
|
|
244
|
+
preprocessMode: "letterbox",
|
|
245
|
+
formats: {
|
|
246
|
+
onnx: {
|
|
247
|
+
url: hf("objectDetection/yolo26/onnx/camstack-yolo26s.onnx"),
|
|
248
|
+
sizeMB: 37
|
|
249
|
+
},
|
|
250
|
+
coreml: {
|
|
251
|
+
url: hf("objectDetection/yolo26/coreml/camstack-yolo26s.mlpackage"),
|
|
252
|
+
sizeMB: 18,
|
|
253
|
+
isDirectory: true,
|
|
254
|
+
files: [...MLPACKAGE_FILES],
|
|
255
|
+
runtimes: ["python"]
|
|
256
|
+
},
|
|
257
|
+
openvino: ovFormat(hf("objectDetection/yolo26/openvino/camstack-yolo26s.xml"), 36)
|
|
258
|
+
}
|
|
259
|
+
},
|
|
260
|
+
{
|
|
261
|
+
id: "yolo26m",
|
|
262
|
+
group: {
|
|
263
|
+
family: "yolo26",
|
|
264
|
+
tier: "m"
|
|
265
|
+
},
|
|
266
|
+
name: "YOLO26 Medium",
|
|
267
|
+
description: "YOLO26 Medium — high accuracy general-purpose detector",
|
|
268
|
+
inputSize: {
|
|
269
|
+
width: 640,
|
|
270
|
+
height: 640
|
|
271
|
+
},
|
|
272
|
+
labels: [],
|
|
273
|
+
preprocessMode: "letterbox",
|
|
274
|
+
formats: {
|
|
275
|
+
onnx: {
|
|
276
|
+
url: hf("objectDetection/yolo26/onnx/camstack-yolo26m.onnx"),
|
|
277
|
+
sizeMB: 78
|
|
278
|
+
},
|
|
279
|
+
coreml: {
|
|
280
|
+
url: hf("objectDetection/yolo26/coreml/camstack-yolo26m.mlpackage"),
|
|
281
|
+
sizeMB: 39,
|
|
282
|
+
isDirectory: true,
|
|
283
|
+
files: [...MLPACKAGE_FILES],
|
|
284
|
+
runtimes: ["python"]
|
|
285
|
+
},
|
|
286
|
+
openvino: ovFormat(hf("objectDetection/yolo26/openvino/camstack-yolo26m.xml"), 78)
|
|
287
|
+
}
|
|
288
|
+
},
|
|
289
|
+
{
|
|
290
|
+
id: "yolo26l",
|
|
291
|
+
group: {
|
|
292
|
+
family: "yolo26",
|
|
293
|
+
tier: "l"
|
|
294
|
+
},
|
|
295
|
+
name: "YOLO26 Large",
|
|
296
|
+
description: "YOLO26 Large — maximum accuracy for demanding deployments",
|
|
297
|
+
inputSize: {
|
|
298
|
+
width: 640,
|
|
299
|
+
height: 640
|
|
300
|
+
},
|
|
301
|
+
labels: [],
|
|
302
|
+
preprocessMode: "letterbox",
|
|
303
|
+
formats: {
|
|
304
|
+
onnx: {
|
|
305
|
+
url: hf("objectDetection/yolo26/onnx/camstack-yolo26l.onnx"),
|
|
306
|
+
sizeMB: 95
|
|
307
|
+
},
|
|
308
|
+
coreml: {
|
|
309
|
+
url: hf("objectDetection/yolo26/coreml/camstack-yolo26l.mlpackage"),
|
|
310
|
+
sizeMB: 48,
|
|
311
|
+
isDirectory: true,
|
|
312
|
+
files: [...MLPACKAGE_FILES],
|
|
313
|
+
runtimes: ["python"]
|
|
314
|
+
},
|
|
315
|
+
openvino: ovFormat(hf("objectDetection/yolo26/openvino/camstack-yolo26l.xml"), 95)
|
|
316
|
+
}
|
|
317
|
+
},
|
|
318
|
+
{
|
|
319
|
+
id: "yolo26x",
|
|
320
|
+
legacy: true,
|
|
321
|
+
name: "YOLO26 XLarge",
|
|
322
|
+
description: "YOLO26 XLarge — highest accuracy, attention-based architecture",
|
|
323
|
+
inputSize: {
|
|
324
|
+
width: 640,
|
|
325
|
+
height: 640
|
|
326
|
+
},
|
|
327
|
+
labels: [],
|
|
328
|
+
preprocessMode: "letterbox",
|
|
329
|
+
formats: {
|
|
330
|
+
onnx: {
|
|
331
|
+
url: hf("objectDetection/yolo26/onnx/camstack-yolo26x.onnx"),
|
|
332
|
+
sizeMB: 213
|
|
333
|
+
},
|
|
334
|
+
coreml: {
|
|
335
|
+
url: hf("objectDetection/yolo26/coreml/camstack-yolo26x.mlpackage"),
|
|
336
|
+
sizeMB: 107,
|
|
337
|
+
isDirectory: true,
|
|
338
|
+
files: [...MLPACKAGE_FILES],
|
|
339
|
+
runtimes: ["python"]
|
|
340
|
+
},
|
|
341
|
+
openvino: ovFormat(hf("objectDetection/yolo26/openvino/camstack-yolo26x.xml"), 213)
|
|
342
|
+
}
|
|
343
|
+
},
|
|
344
|
+
{
|
|
345
|
+
id: "scrypted-yolov9t-relu",
|
|
346
|
+
legacy: true,
|
|
347
|
+
name: "YOLOv9 Tiny ReLU",
|
|
348
|
+
description: "YOLOv9t ReLU — optimized for int8 quantization",
|
|
349
|
+
inputSize: {
|
|
350
|
+
width: 320,
|
|
351
|
+
height: 320
|
|
352
|
+
},
|
|
353
|
+
labels: [],
|
|
354
|
+
preprocessMode: "letterbox",
|
|
355
|
+
formats: {
|
|
356
|
+
onnx: {
|
|
357
|
+
url: hfScrypted("onnx/scrypted_yolov9t_relu/scrypted_yolov9t_relu.onnx"),
|
|
358
|
+
sizeMB: 8.1
|
|
359
|
+
},
|
|
360
|
+
coreml: {
|
|
361
|
+
url: hfScrypted("coreml/scrypted_yolov9t_relu/scrypted_yolov9t_relu.mlpackage"),
|
|
362
|
+
sizeMB: 4,
|
|
363
|
+
isDirectory: true,
|
|
364
|
+
files: [...MLPACKAGE_FILES],
|
|
365
|
+
runtimes: ["python"]
|
|
366
|
+
},
|
|
367
|
+
openvino: ovFormat(hf("objectDetection/scrypted-yolov9-relu/openvino/scrypted_yolov9t_relu.xml"), 6)
|
|
368
|
+
}
|
|
369
|
+
},
|
|
370
|
+
{
|
|
371
|
+
id: "scrypted-yolov9s-relu",
|
|
372
|
+
legacy: true,
|
|
373
|
+
name: "YOLOv9 Small ReLU",
|
|
374
|
+
description: "YOLOv9s ReLU — improved efficiency, int8 ready",
|
|
375
|
+
inputSize: {
|
|
376
|
+
width: 320,
|
|
377
|
+
height: 320
|
|
378
|
+
},
|
|
379
|
+
labels: [],
|
|
380
|
+
preprocessMode: "letterbox",
|
|
381
|
+
formats: {
|
|
382
|
+
onnx: {
|
|
383
|
+
url: hfScrypted("onnx/scrypted_yolov9s_relu/scrypted_yolov9s_relu.onnx"),
|
|
384
|
+
sizeMB: 27.9
|
|
385
|
+
},
|
|
386
|
+
coreml: {
|
|
387
|
+
url: hfScrypted("coreml/scrypted_yolov9s_relu/scrypted_yolov9s_relu.mlpackage"),
|
|
388
|
+
sizeMB: 14,
|
|
389
|
+
isDirectory: true,
|
|
390
|
+
files: [...MLPACKAGE_FILES],
|
|
391
|
+
runtimes: ["python"]
|
|
392
|
+
},
|
|
393
|
+
openvino: ovFormat(hf("objectDetection/scrypted-yolov9-relu/openvino/scrypted_yolov9s_relu.xml"), 16)
|
|
394
|
+
}
|
|
395
|
+
},
|
|
396
|
+
{
|
|
397
|
+
id: "scrypted-yolov9c-relu",
|
|
398
|
+
legacy: true,
|
|
399
|
+
name: "YOLOv9 C ReLU",
|
|
400
|
+
description: "YOLOv9c ReLU — high-accuracy, int8 ready",
|
|
401
|
+
inputSize: {
|
|
402
|
+
width: 320,
|
|
403
|
+
height: 320
|
|
404
|
+
},
|
|
405
|
+
labels: [],
|
|
406
|
+
preprocessMode: "letterbox",
|
|
407
|
+
formats: {
|
|
408
|
+
onnx: {
|
|
409
|
+
url: hfScrypted("onnx/scrypted_yolov9c_relu/scrypted_yolov9c_relu.onnx"),
|
|
410
|
+
sizeMB: 96.9
|
|
411
|
+
},
|
|
412
|
+
coreml: {
|
|
413
|
+
url: hfScrypted("coreml/scrypted_yolov9c_relu/scrypted_yolov9c_relu.mlpackage"),
|
|
414
|
+
sizeMB: 48,
|
|
415
|
+
isDirectory: true,
|
|
416
|
+
files: [...MLPACKAGE_FILES],
|
|
417
|
+
runtimes: ["python"]
|
|
418
|
+
},
|
|
419
|
+
openvino: ovFormat(hf("objectDetection/scrypted-yolov9-relu/openvino/scrypted_yolov9c_relu.xml"), 49)
|
|
420
|
+
}
|
|
421
|
+
},
|
|
422
|
+
{
|
|
423
|
+
id: "scrypted-yolov9m-relu",
|
|
424
|
+
legacy: true,
|
|
425
|
+
name: "YOLOv9 M ReLU",
|
|
426
|
+
description: "YOLOv9m ReLU — medium, int8 ready",
|
|
427
|
+
inputSize: {
|
|
428
|
+
width: 320,
|
|
429
|
+
height: 320
|
|
430
|
+
},
|
|
431
|
+
labels: [],
|
|
432
|
+
preprocessMode: "letterbox",
|
|
433
|
+
formats: {
|
|
434
|
+
onnx: {
|
|
435
|
+
url: hfScrypted("onnx/scrypted_yolov9m_relu/scrypted_yolov9m_relu.onnx"),
|
|
436
|
+
sizeMB: 76.6
|
|
437
|
+
},
|
|
438
|
+
coreml: {
|
|
439
|
+
url: hfScrypted("coreml/scrypted_yolov9m_relu/scrypted_yolov9m_relu.mlpackage"),
|
|
440
|
+
sizeMB: 38,
|
|
441
|
+
isDirectory: true,
|
|
442
|
+
files: [...MLPACKAGE_FILES],
|
|
443
|
+
runtimes: ["python"]
|
|
444
|
+
},
|
|
445
|
+
openvino: ovFormat(hf("objectDetection/scrypted-yolov9-relu/openvino/scrypted_yolov9m_relu.xml"), 38)
|
|
446
|
+
}
|
|
447
|
+
},
|
|
448
|
+
ovPrecisionVariant("yolov9t", "objectDetection/yolov9/openvino", "YOLOv9 Tiny", "fp16", 5, true),
|
|
449
|
+
ovPrecisionVariant("yolov9t", "objectDetection/yolov9/openvino", "YOLOv9 Tiny", "int8", 3, true),
|
|
450
|
+
ovPrecisionVariant("yolov9s", "objectDetection/yolov9/openvino", "YOLOv9 Small", "fp16", 15, true),
|
|
451
|
+
ovPrecisionVariant("yolov9s", "objectDetection/yolov9/openvino", "YOLOv9 Small", "int8", 8, true),
|
|
452
|
+
ovPrecisionVariant("yolo26n", "objectDetection/yolo26/openvino", "YOLO26 Nano", "fp16", 5, true),
|
|
453
|
+
ovPrecisionVariant("yolo26n", "objectDetection/yolo26/openvino", "YOLO26 Nano", "int8", 3),
|
|
454
|
+
ovPrecisionVariant("yolo26s", "objectDetection/yolo26/openvino", "YOLO26 Small", "fp16", 19, true),
|
|
455
|
+
ovPrecisionVariant("yolo26s", "objectDetection/yolo26/openvino", "YOLO26 Small", "int8", 10),
|
|
456
|
+
ovPrecisionVariant("yolo26m", "objectDetection/yolo26/openvino", "YOLO26 Medium", "fp16", 41, true),
|
|
457
|
+
ovPrecisionVariant("yolo26m", "objectDetection/yolo26/openvino", "YOLO26 Medium", "int8", 21),
|
|
458
|
+
ovPrecisionVariant("yolo26l", "objectDetection/yolo26/openvino", "YOLO26 Large", "fp16", 50, true),
|
|
459
|
+
ovPrecisionVariant("yolo26l", "objectDetection/yolo26/openvino", "YOLO26 Large", "int8", 25),
|
|
460
|
+
...yolo26Reduced("n", 320, {
|
|
461
|
+
onnx: 10.47,
|
|
462
|
+
coreml: 5.05,
|
|
463
|
+
ovFp32: 10.05,
|
|
464
|
+
ovInt8: 3.29
|
|
465
|
+
}),
|
|
466
|
+
...yolo26Reduced("n", 256, {
|
|
467
|
+
onnx: 10.47,
|
|
468
|
+
coreml: 5.05,
|
|
469
|
+
ovFp32: 10.05,
|
|
470
|
+
ovInt8: 3.29
|
|
471
|
+
}),
|
|
472
|
+
...yolo26Reduced("s", 320, {
|
|
473
|
+
onnx: 38.82,
|
|
474
|
+
coreml: 19.23,
|
|
475
|
+
ovFp32: 38.4,
|
|
476
|
+
ovInt8: 10.43
|
|
477
|
+
}),
|
|
478
|
+
...yolo26Reduced("s", 256, {
|
|
479
|
+
onnx: 38.82,
|
|
480
|
+
coreml: 19.23,
|
|
481
|
+
ovFp32: 38.4,
|
|
482
|
+
ovInt8: 10.43
|
|
483
|
+
}),
|
|
484
|
+
...yolo26Reduced("m", 320, {
|
|
485
|
+
onnx: 83.31,
|
|
486
|
+
coreml: 41.07,
|
|
487
|
+
ovFp32: 82.09,
|
|
488
|
+
ovInt8: 21.47
|
|
489
|
+
}),
|
|
490
|
+
...yolo26Reduced("m", 256, {
|
|
491
|
+
onnx: 83.31,
|
|
492
|
+
coreml: 41.07,
|
|
493
|
+
ovFp32: 82.09,
|
|
494
|
+
ovInt8: 21.47
|
|
495
|
+
}),
|
|
496
|
+
...yolo26Reduced("l", 320, {
|
|
497
|
+
onnx: 101.74,
|
|
498
|
+
coreml: 49.97,
|
|
499
|
+
ovFp32: 99.85,
|
|
500
|
+
ovInt8: 26.31
|
|
501
|
+
}),
|
|
502
|
+
...yolo26Reduced("l", 256, {
|
|
503
|
+
onnx: 101.74,
|
|
504
|
+
coreml: 49.97,
|
|
505
|
+
ovFp32: 99.85,
|
|
506
|
+
ovInt8: 26.31
|
|
507
|
+
}),
|
|
508
|
+
ovPrecisionVariant("yolo26x", "objectDetection/yolo26/openvino", "YOLO26 XLarge", "fp16", 112, true),
|
|
509
|
+
ovPrecisionVariant("yolo26x", "objectDetection/yolo26/openvino", "YOLO26 XLarge", "int8", 56, true)
|
|
510
|
+
];
|
|
511
|
+
var FACE_DETECTION_MODELS = [{
|
|
512
|
+
id: "scrfd-2.5g",
|
|
513
|
+
name: "SCRFD 2.5G",
|
|
514
|
+
description: "SCRFD 2.5G — balanced face detection model",
|
|
515
|
+
inputSize: {
|
|
516
|
+
width: 640,
|
|
517
|
+
height: 640
|
|
518
|
+
},
|
|
519
|
+
labels: [{
|
|
520
|
+
id: "face",
|
|
521
|
+
name: "Face"
|
|
522
|
+
}],
|
|
523
|
+
preprocessMode: "letterbox",
|
|
524
|
+
formats: {
|
|
525
|
+
onnx: {
|
|
526
|
+
url: hf("faceDetection/scrfd/onnx/camstack-scrfd-2.5g.onnx"),
|
|
527
|
+
sizeMB: 3.1
|
|
528
|
+
},
|
|
529
|
+
coreml: {
|
|
530
|
+
url: hf("faceDetection/scrfd/coreml/camstack-scrfd-2.5g.mlpackage"),
|
|
531
|
+
sizeMB: 1.7,
|
|
532
|
+
isDirectory: true,
|
|
533
|
+
files: [...MLPACKAGE_FILES],
|
|
534
|
+
runtimes: ["python"]
|
|
535
|
+
},
|
|
536
|
+
openvino: ovFormat(hf("faceDetection/scrfd/openvino/camstack-scrfd-2.5g.xml"), 1.8)
|
|
537
|
+
}
|
|
538
|
+
}, {
|
|
539
|
+
id: "scrypted-yolov9t-face",
|
|
540
|
+
name: "YOLOv9t Face ReLU",
|
|
541
|
+
description: "YOLOv9t face detection — YOLO-based, fast, 320x320 ReLU",
|
|
542
|
+
inputSize: {
|
|
543
|
+
width: 320,
|
|
544
|
+
height: 320
|
|
545
|
+
},
|
|
546
|
+
labels: [{
|
|
547
|
+
id: "face",
|
|
548
|
+
name: "Face"
|
|
549
|
+
}],
|
|
550
|
+
preprocessMode: "letterbox",
|
|
551
|
+
formats: {
|
|
552
|
+
onnx: {
|
|
553
|
+
url: hfScrypted("onnx/scrypted_yolov9t_relu_face/scrypted_yolov9t_relu_face.onnx"),
|
|
554
|
+
sizeMB: 8.1
|
|
555
|
+
},
|
|
556
|
+
coreml: {
|
|
557
|
+
url: hfScrypted("coreml/scrypted_yolov9t_relu_face/scrypted_yolov9t_relu_face.mlpackage"),
|
|
558
|
+
sizeMB: 4,
|
|
559
|
+
isDirectory: true,
|
|
560
|
+
files: [...MLPACKAGE_FILES],
|
|
561
|
+
runtimes: ["python"]
|
|
562
|
+
},
|
|
563
|
+
openvino: ovFormat(hf("faceDetection/scrypted-yolov9-face/openvino/scrypted_yolov9t_relu_face.xml"), 6)
|
|
564
|
+
}
|
|
565
|
+
}];
|
|
566
|
+
var FACE_EMBEDDING_MODELS = [{
|
|
567
|
+
id: "arcface-r100",
|
|
568
|
+
name: "ArcFace R100",
|
|
569
|
+
description: "ArcFace ResNet-100 — high-accuracy face recognition embeddings (512-d)",
|
|
570
|
+
inputSize: {
|
|
571
|
+
width: 112,
|
|
572
|
+
height: 112
|
|
573
|
+
},
|
|
574
|
+
inputLayout: "nhwc",
|
|
575
|
+
labels: [{
|
|
576
|
+
id: "embedding",
|
|
577
|
+
name: "Face Embedding"
|
|
578
|
+
}],
|
|
579
|
+
preprocessMode: "resize",
|
|
580
|
+
faceAlignment: true,
|
|
581
|
+
formats: {
|
|
582
|
+
onnx: {
|
|
583
|
+
url: hf("faceRecognition/arcface/onnx/camstack-arcface-arcface.onnx"),
|
|
584
|
+
sizeMB: 130
|
|
585
|
+
},
|
|
586
|
+
coreml: {
|
|
587
|
+
url: hf("faceRecognition/arcface/coreml/camstack-arcface-r100.mlpackage"),
|
|
588
|
+
sizeMB: 65,
|
|
589
|
+
isDirectory: true,
|
|
590
|
+
files: [...MLPACKAGE_FILES],
|
|
591
|
+
runtimes: ["python"]
|
|
592
|
+
},
|
|
593
|
+
openvino: ovFormat(hf("faceRecognition/arcface/openvino/camstack-arcface-r100.xml"), 65)
|
|
594
|
+
}
|
|
595
|
+
}, {
|
|
596
|
+
id: "inception-resnet-v1",
|
|
597
|
+
name: "Inception ResNet V1",
|
|
598
|
+
description: "FaceNet-style face recognition embeddings (512-d) — hosted on plugin-models HF repo",
|
|
599
|
+
inputSize: {
|
|
600
|
+
width: 160,
|
|
601
|
+
height: 160
|
|
602
|
+
},
|
|
603
|
+
labels: [{
|
|
604
|
+
id: "embedding",
|
|
605
|
+
name: "Face Embedding"
|
|
606
|
+
}],
|
|
607
|
+
preprocessMode: "resize",
|
|
608
|
+
formats: {
|
|
609
|
+
onnx: {
|
|
610
|
+
url: hfScrypted("onnx/inception_resnet_v1/inception_resnet_v1.onnx"),
|
|
611
|
+
sizeMB: 89.6
|
|
612
|
+
},
|
|
613
|
+
coreml: {
|
|
614
|
+
url: hfScrypted("coreml/inception_resnet_v1/inception_resnet_v1.mlpackage"),
|
|
615
|
+
sizeMB: 45,
|
|
616
|
+
isDirectory: true,
|
|
617
|
+
files: [...MLPACKAGE_FILES],
|
|
618
|
+
runtimes: ["python"]
|
|
619
|
+
},
|
|
620
|
+
openvino: ovFormat(hf("faceRecognition/inception-resnet-v1/openvino/camstack-inception-resnet-v1.xml"), 45)
|
|
621
|
+
}
|
|
622
|
+
}];
|
|
623
|
+
var PLATE_DETECTION_MODELS = [{
|
|
624
|
+
id: "yolov8n-plate",
|
|
625
|
+
name: "YOLOv8 Nano — License Plate",
|
|
626
|
+
description: "YOLOv8 Nano fine-tuned for license plate detection",
|
|
627
|
+
inputSize: {
|
|
628
|
+
width: 640,
|
|
629
|
+
height: 640
|
|
630
|
+
},
|
|
631
|
+
labels: [{
|
|
632
|
+
id: "plate",
|
|
633
|
+
name: "License Plate"
|
|
634
|
+
}],
|
|
635
|
+
preprocessMode: "letterbox",
|
|
636
|
+
formats: {
|
|
637
|
+
onnx: {
|
|
638
|
+
url: hf("plateDetection/yolov8-plate/onnx/camstack-yolov8n-plate.onnx"),
|
|
639
|
+
sizeMB: 12
|
|
640
|
+
},
|
|
641
|
+
coreml: {
|
|
642
|
+
url: hf("plateDetection/yolov8-plate/coreml/camstack-yolov8n-plate.mlpackage"),
|
|
643
|
+
sizeMB: 5.9,
|
|
644
|
+
isDirectory: true,
|
|
645
|
+
files: [...MLPACKAGE_FILES],
|
|
646
|
+
runtimes: ["python"]
|
|
647
|
+
},
|
|
648
|
+
openvino: ovFormat(hf("plateDetection/yolov8-plate/openvino/camstack-yolov8n-plate.xml"), 6.1)
|
|
649
|
+
}
|
|
650
|
+
}];
|
|
651
|
+
var PLATE_OCR_MODELS = [{
|
|
652
|
+
id: "vgg-english-g2",
|
|
653
|
+
name: "VGG English G2",
|
|
654
|
+
description: "EasyOCR VGG English G2 — text recognition for license plates and scene text",
|
|
655
|
+
inputSize: {
|
|
656
|
+
width: 384,
|
|
657
|
+
height: 64
|
|
658
|
+
},
|
|
659
|
+
labels: [{
|
|
660
|
+
id: "text",
|
|
661
|
+
name: "Plate Text"
|
|
662
|
+
}],
|
|
663
|
+
preprocessMode: "resize",
|
|
664
|
+
formats: {
|
|
665
|
+
onnx: {
|
|
666
|
+
url: hfScrypted("onnx/vgg_english_g2/vgg_english_g2.onnx"),
|
|
667
|
+
sizeMB: 14.4
|
|
668
|
+
},
|
|
669
|
+
coreml: {
|
|
670
|
+
url: hfScrypted("coreml/vgg_english_g2/vgg_english_g2.mlpackage"),
|
|
671
|
+
sizeMB: 11.2,
|
|
672
|
+
isDirectory: true,
|
|
673
|
+
files: [...MLPACKAGE_FILES],
|
|
674
|
+
runtimes: ["python"]
|
|
675
|
+
},
|
|
676
|
+
openvino: ovFormat(hf("plateRecognition/vgg_english_g2/openvino/vgg_english_g2.xml"), 7.2)
|
|
677
|
+
}
|
|
678
|
+
}];
|
|
679
|
+
var ANIMAL_CLASSIFIER_MODELS = [{
|
|
680
|
+
id: "animals-10",
|
|
681
|
+
name: "Animal Classifier (10)",
|
|
682
|
+
description: "ViT-based animal type classifier — cat, cow, dog, dolphin, eagle, panda, horse, monkey, sheep, spider",
|
|
683
|
+
inputSize: {
|
|
684
|
+
width: 224,
|
|
685
|
+
height: 224
|
|
686
|
+
},
|
|
687
|
+
inputNormalization: "imagenet",
|
|
688
|
+
labels: [{
|
|
689
|
+
id: "animal-type",
|
|
690
|
+
name: "Animal Type"
|
|
691
|
+
}],
|
|
692
|
+
preprocessMode: "resize",
|
|
693
|
+
formats: {
|
|
694
|
+
onnx: {
|
|
695
|
+
url: hf("animalClassification/animals-10/onnx/camstack-animals-10.onnx"),
|
|
696
|
+
sizeMB: 328
|
|
697
|
+
},
|
|
698
|
+
coreml: {
|
|
699
|
+
url: hf("animalClassification/animals-10/coreml/camstack-animals-10.mlpackage"),
|
|
700
|
+
sizeMB: 164,
|
|
701
|
+
isDirectory: true,
|
|
702
|
+
files: [...MLPACKAGE_FILES],
|
|
703
|
+
runtimes: ["python"]
|
|
704
|
+
},
|
|
705
|
+
openvino: ovFormat(hf("animalClassification/animals-10/openvino/camstack-animals-10.xml"), 164)
|
|
706
|
+
}
|
|
707
|
+
}];
|
|
708
|
+
var BIRD_CLASSIFIER_MODELS = [{
|
|
709
|
+
id: "bird-nabirds-404",
|
|
710
|
+
name: "NABirds (404 species)",
|
|
711
|
+
description: "ResNet50 trained on NABirds — 404 North American species",
|
|
712
|
+
inputSize: {
|
|
713
|
+
width: 224,
|
|
714
|
+
height: 224
|
|
715
|
+
},
|
|
716
|
+
inputNormalization: "imagenet",
|
|
717
|
+
labels: [{
|
|
718
|
+
id: "species",
|
|
719
|
+
name: "Bird Species"
|
|
720
|
+
}],
|
|
721
|
+
preprocessMode: "resize",
|
|
722
|
+
formats: {
|
|
723
|
+
onnx: {
|
|
724
|
+
url: hf("animalClassification/bird-nabirds/onnx/camstack-bird-nabirds-404.onnx"),
|
|
725
|
+
sizeMB: 93
|
|
726
|
+
},
|
|
727
|
+
coreml: {
|
|
728
|
+
url: hf("animalClassification/bird-nabirds/coreml/camstack-bird-nabirds-404.mlpackage"),
|
|
729
|
+
sizeMB: 47,
|
|
730
|
+
isDirectory: true,
|
|
731
|
+
files: [...MLPACKAGE_FILES],
|
|
732
|
+
runtimes: ["python"]
|
|
733
|
+
},
|
|
734
|
+
openvino: ovFormat(hf("animalClassification/bird-nabirds/openvino/camstack-bird-nabirds-404.xml"), 47)
|
|
735
|
+
},
|
|
736
|
+
extraFiles: [{
|
|
737
|
+
url: hf("animalClassification/bird-nabirds/onnx/camstack-bird-nabirds-404-labels.json"),
|
|
738
|
+
filename: "camstack-bird-nabirds-404-labels.json",
|
|
739
|
+
sizeMB: .02
|
|
740
|
+
}]
|
|
741
|
+
}];
|
|
742
|
+
var VEHICLE_CLASSIFIER_MODELS = [{
|
|
743
|
+
id: "vehicle-type-efficientnet",
|
|
744
|
+
name: "Vehicle Type (EfficientNet)",
|
|
745
|
+
description: "EfficientNet-B4 vehicle make/model/year classifier — 8,949 classes from VMMRdb",
|
|
746
|
+
inputSize: {
|
|
747
|
+
width: 380,
|
|
748
|
+
height: 380
|
|
749
|
+
},
|
|
750
|
+
inputNormalization: "imagenet",
|
|
751
|
+
labels: [{
|
|
752
|
+
id: "vehicle-type",
|
|
753
|
+
name: "Vehicle Type"
|
|
754
|
+
}],
|
|
755
|
+
preprocessMode: "resize",
|
|
756
|
+
formats: {
|
|
757
|
+
onnx: {
|
|
758
|
+
url: hf("vehicleClassification/efficientnet/onnx/camstack-vehicle-type-efficientnet.onnx"),
|
|
759
|
+
sizeMB: 135
|
|
760
|
+
},
|
|
761
|
+
coreml: {
|
|
762
|
+
url: hf("vehicleClassification/efficientnet/coreml/camstack-vehicle-type-efficientnet.mlpackage"),
|
|
763
|
+
sizeMB: 10,
|
|
764
|
+
isDirectory: true,
|
|
765
|
+
files: [...MLPACKAGE_FILES],
|
|
766
|
+
runtimes: ["python"]
|
|
767
|
+
},
|
|
768
|
+
openvino: ovFormat(hf("vehicleClassification/efficientnet/openvino/camstack-vehicle-type-efficientnet.xml"), 68)
|
|
769
|
+
},
|
|
770
|
+
extraFiles: [{
|
|
771
|
+
url: hf("vehicleClassification/efficientnet/camstack-vehicle-type-labels.json"),
|
|
772
|
+
filename: "camstack-vehicle-type-labels.json",
|
|
773
|
+
sizeMB: .2
|
|
774
|
+
}]
|
|
775
|
+
}];
|
|
776
|
+
var SEGMENTATION_REFINER_MODELS = [{
|
|
777
|
+
id: "u2netp",
|
|
778
|
+
name: "U2-Net Portable",
|
|
779
|
+
description: "U2-Net-P — ultra-lightweight salient object segmentation (4.7 MB)",
|
|
780
|
+
inputSize: {
|
|
781
|
+
width: 320,
|
|
782
|
+
height: 320
|
|
783
|
+
},
|
|
784
|
+
labels: [{
|
|
785
|
+
id: "mask",
|
|
786
|
+
name: "Segmentation Mask"
|
|
787
|
+
}],
|
|
788
|
+
preprocessMode: "resize",
|
|
789
|
+
formats: {
|
|
790
|
+
onnx: {
|
|
791
|
+
url: hf("segmentationRefiner/u2netp/onnx/camstack-u2netp.onnx"),
|
|
792
|
+
sizeMB: 5
|
|
793
|
+
},
|
|
794
|
+
coreml: {
|
|
795
|
+
url: hf("segmentationRefiner/u2netp/coreml/camstack-u2netp.mlpackage"),
|
|
796
|
+
sizeMB: 3,
|
|
797
|
+
isDirectory: true,
|
|
798
|
+
files: [...MLPACKAGE_FILES],
|
|
799
|
+
runtimes: ["python"]
|
|
800
|
+
},
|
|
801
|
+
openvino: ovFormat(hf("segmentationRefiner/u2netp/openvino/camstack-u2netp.xml"), 2.5)
|
|
802
|
+
}
|
|
803
|
+
}];
|
|
804
|
+
var INSTANCE_SEGMENTATION_MODELS = [
|
|
805
|
+
{
|
|
806
|
+
id: "yolo26n-seg",
|
|
807
|
+
name: "YOLO26 Nano Seg",
|
|
808
|
+
description: "YOLO26 Nano Segmentation — ultra-lightweight instance segmentation with masks",
|
|
809
|
+
inputSize: {
|
|
810
|
+
width: 640,
|
|
811
|
+
height: 640
|
|
812
|
+
},
|
|
813
|
+
labels: [],
|
|
814
|
+
preprocessMode: "letterbox",
|
|
815
|
+
formats: {
|
|
816
|
+
onnx: {
|
|
817
|
+
url: hf("segmentation/yolo26-seg/onnx/camstack-yolo26n-seg.onnx"),
|
|
818
|
+
sizeMB: 11
|
|
819
|
+
},
|
|
820
|
+
coreml: {
|
|
821
|
+
url: hf("segmentation/yolo26-seg/coreml/camstack-yolo26n-seg.mlpackage"),
|
|
822
|
+
sizeMB: 6,
|
|
823
|
+
isDirectory: true,
|
|
824
|
+
files: [...MLPACKAGE_FILES],
|
|
825
|
+
runtimes: ["python"]
|
|
826
|
+
},
|
|
827
|
+
openvino: ovFormat(hf("segmentation/yolo26-seg/openvino/camstack-yolo26n-seg.xml"), 11)
|
|
828
|
+
}
|
|
829
|
+
},
|
|
830
|
+
{
|
|
831
|
+
id: "yolo26s-seg",
|
|
832
|
+
name: "YOLO26 Small Seg",
|
|
833
|
+
description: "YOLO26 Small Segmentation — balanced instance segmentation",
|
|
834
|
+
inputSize: {
|
|
835
|
+
width: 640,
|
|
836
|
+
height: 640
|
|
837
|
+
},
|
|
838
|
+
labels: [],
|
|
839
|
+
preprocessMode: "letterbox",
|
|
840
|
+
formats: {
|
|
841
|
+
onnx: {
|
|
842
|
+
url: hf("segmentation/yolo26-seg/onnx/camstack-yolo26s-seg.onnx"),
|
|
843
|
+
sizeMB: 40
|
|
844
|
+
},
|
|
845
|
+
coreml: {
|
|
846
|
+
url: hf("segmentation/yolo26-seg/coreml/camstack-yolo26s-seg.mlpackage"),
|
|
847
|
+
sizeMB: 20,
|
|
848
|
+
isDirectory: true,
|
|
849
|
+
files: [...MLPACKAGE_FILES],
|
|
850
|
+
runtimes: ["python"]
|
|
851
|
+
},
|
|
852
|
+
openvino: ovFormat(hf("segmentation/yolo26-seg/openvino/camstack-yolo26s-seg.xml"), 40)
|
|
853
|
+
}
|
|
854
|
+
},
|
|
855
|
+
{
|
|
856
|
+
id: "yolo26m-seg",
|
|
857
|
+
name: "YOLO26 Medium Seg",
|
|
858
|
+
description: "YOLO26 Medium Segmentation — high-accuracy instance segmentation",
|
|
859
|
+
inputSize: {
|
|
860
|
+
width: 640,
|
|
861
|
+
height: 640
|
|
862
|
+
},
|
|
863
|
+
labels: [],
|
|
864
|
+
preprocessMode: "letterbox",
|
|
865
|
+
formats: {
|
|
866
|
+
onnx: {
|
|
867
|
+
url: hf("segmentation/yolo26-seg/onnx/camstack-yolo26m-seg.onnx"),
|
|
868
|
+
sizeMB: 90
|
|
869
|
+
},
|
|
870
|
+
coreml: {
|
|
871
|
+
url: hf("segmentation/yolo26-seg/coreml/camstack-yolo26m-seg.mlpackage"),
|
|
872
|
+
sizeMB: 45,
|
|
873
|
+
isDirectory: true,
|
|
874
|
+
files: [...MLPACKAGE_FILES],
|
|
875
|
+
runtimes: ["python"]
|
|
876
|
+
},
|
|
877
|
+
openvino: ovFormat(hf("segmentation/yolo26-seg/openvino/camstack-yolo26m-seg.xml"), 90)
|
|
878
|
+
}
|
|
879
|
+
}
|
|
880
|
+
];
|
|
881
|
+
var CLIP_EMBEDDING_MODELS = [
|
|
882
|
+
{
|
|
883
|
+
id: "mobileclip-s0",
|
|
884
|
+
name: "MobileCLIP S0",
|
|
885
|
+
description: "MobileCLIP S0 — Apple ultra-fast CLIP vision encoder, 512-dim, 256×256 (40 MB ONNX)",
|
|
886
|
+
inputSize: {
|
|
887
|
+
width: 256,
|
|
888
|
+
height: 256
|
|
889
|
+
},
|
|
890
|
+
labels: [{
|
|
891
|
+
id: "embedding",
|
|
892
|
+
name: "CLIP Embedding"
|
|
893
|
+
}],
|
|
894
|
+
preprocessMode: "resize",
|
|
895
|
+
inputNormalization: "none",
|
|
896
|
+
formats: {
|
|
897
|
+
onnx: {
|
|
898
|
+
url: hf("clip/mobileclip-s0/onnx/camstack-mobileclip-s0-vision.onnx"),
|
|
899
|
+
sizeMB: 40
|
|
900
|
+
},
|
|
901
|
+
openvino: ovFormat(hf("clip/mobileclip-s0/openvino/camstack-mobileclip-s0-vision.xml"), 25),
|
|
902
|
+
coreml: {
|
|
903
|
+
url: hf("clip/mobileclip-s0/coreml/camstack-mobileclip-s0-vision.mlpackage"),
|
|
904
|
+
sizeMB: 30,
|
|
905
|
+
isDirectory: true,
|
|
906
|
+
files: [...MLPACKAGE_FILES],
|
|
907
|
+
runtimes: ["python"]
|
|
908
|
+
}
|
|
909
|
+
}
|
|
910
|
+
},
|
|
911
|
+
{
|
|
912
|
+
id: "mobileclip-s1",
|
|
913
|
+
name: "MobileCLIP S1",
|
|
914
|
+
description: "MobileCLIP S1 — Apple balanced CLIP vision encoder, 512-dim, 256×256 (90 MB ONNX)",
|
|
915
|
+
inputSize: {
|
|
916
|
+
width: 256,
|
|
917
|
+
height: 256
|
|
918
|
+
},
|
|
919
|
+
labels: [{
|
|
920
|
+
id: "embedding",
|
|
921
|
+
name: "CLIP Embedding"
|
|
922
|
+
}],
|
|
923
|
+
preprocessMode: "resize",
|
|
924
|
+
inputNormalization: "none",
|
|
925
|
+
formats: {
|
|
926
|
+
onnx: {
|
|
927
|
+
url: hf("clip/mobileclip-s1/onnx/camstack-mobileclip-s1-vision.onnx"),
|
|
928
|
+
sizeMB: 90
|
|
929
|
+
},
|
|
930
|
+
openvino: ovFormat(hf("clip/mobileclip-s1/openvino/camstack-mobileclip-s1-vision.xml"), 55),
|
|
931
|
+
coreml: {
|
|
932
|
+
url: hf("clip/mobileclip-s1/coreml/camstack-mobileclip-s1-vision.mlpackage"),
|
|
933
|
+
sizeMB: 65,
|
|
934
|
+
isDirectory: true,
|
|
935
|
+
files: [...MLPACKAGE_FILES],
|
|
936
|
+
runtimes: ["python"]
|
|
937
|
+
}
|
|
938
|
+
}
|
|
939
|
+
},
|
|
940
|
+
{
|
|
941
|
+
id: "mobileclip-s2",
|
|
942
|
+
name: "MobileCLIP S2",
|
|
943
|
+
description: "MobileCLIP S2 — Apple high-accuracy CLIP vision encoder, 512-dim, 256×256 (150 MB ONNX)",
|
|
944
|
+
inputSize: {
|
|
945
|
+
width: 256,
|
|
946
|
+
height: 256
|
|
947
|
+
},
|
|
948
|
+
labels: [{
|
|
949
|
+
id: "embedding",
|
|
950
|
+
name: "CLIP Embedding"
|
|
951
|
+
}],
|
|
952
|
+
preprocessMode: "resize",
|
|
953
|
+
inputNormalization: "none",
|
|
954
|
+
formats: {
|
|
955
|
+
onnx: {
|
|
956
|
+
url: hf("clip/mobileclip-s2/onnx/camstack-mobileclip-s2-vision.onnx"),
|
|
957
|
+
sizeMB: 150
|
|
958
|
+
},
|
|
959
|
+
openvino: ovFormat(hf("clip/mobileclip-s2/openvino/camstack-mobileclip-s2-vision.xml"), 90),
|
|
960
|
+
coreml: {
|
|
961
|
+
url: hf("clip/mobileclip-s2/coreml/camstack-mobileclip-s2-vision.mlpackage"),
|
|
962
|
+
sizeMB: 110,
|
|
963
|
+
isDirectory: true,
|
|
964
|
+
files: [...MLPACKAGE_FILES],
|
|
965
|
+
runtimes: ["python"]
|
|
966
|
+
}
|
|
967
|
+
}
|
|
968
|
+
}
|
|
969
|
+
];
|
|
970
|
+
var AUDIO_CLASSIFIER_MODELS = [{
|
|
971
|
+
id: "yamnet-onnx",
|
|
972
|
+
name: "YAMNet",
|
|
973
|
+
description: "Google YAMNet — 521-class audio event classifier (3.2 MB ONNX, runs on any platform)",
|
|
974
|
+
inputSize: {
|
|
975
|
+
width: 1,
|
|
976
|
+
height: 16e3
|
|
977
|
+
},
|
|
978
|
+
labels: [],
|
|
979
|
+
preprocessMode: "resize",
|
|
980
|
+
formats: {
|
|
981
|
+
onnx: {
|
|
982
|
+
url: hf("audioClassification/yamnet/onnx/camstack-yamnet.onnx"),
|
|
983
|
+
sizeMB: 3.2
|
|
984
|
+
},
|
|
985
|
+
openvino: ovFormat(hf("audioClassification/yamnet/openvino/camstack-yamnet.xml"), 3.2),
|
|
986
|
+
coreml: {
|
|
987
|
+
url: hf("audioClassification/yamnet/onnx/camstack-yamnet.onnx"),
|
|
988
|
+
sizeMB: 3.2,
|
|
989
|
+
runtimes: ["python"]
|
|
990
|
+
}
|
|
991
|
+
}
|
|
992
|
+
}, {
|
|
993
|
+
id: "apple-soundanalysis",
|
|
994
|
+
name: "Apple SoundAnalysis",
|
|
995
|
+
description: "macOS built-in — 303 sound categories, Neural Engine accelerated, zero download",
|
|
996
|
+
inputSize: {
|
|
997
|
+
width: 1,
|
|
998
|
+
height: 16e3
|
|
999
|
+
},
|
|
1000
|
+
labels: [],
|
|
1001
|
+
preprocessMode: "resize",
|
|
1002
|
+
formats: { coreml: {
|
|
1003
|
+
url: "",
|
|
1004
|
+
sizeMB: 0
|
|
1005
|
+
} }
|
|
1006
|
+
}];
|
|
1007
|
+
//#endregion
|
|
1008
|
+
//#region src/detection-pipeline/registry/step-definitions.ts
|
|
1009
|
+
var ANIMAL_10_LABELS = [
|
|
1010
|
+
"cat",
|
|
1011
|
+
"cow",
|
|
1012
|
+
"dog",
|
|
1013
|
+
"dolphin",
|
|
1014
|
+
"eagle",
|
|
1015
|
+
"giant panda",
|
|
1016
|
+
"horse",
|
|
1017
|
+
"monkey",
|
|
1018
|
+
"sheep",
|
|
1019
|
+
"spider"
|
|
1020
|
+
];
|
|
1021
|
+
var VGG_G2_CHARSET = ["", ..."0123456789!\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~ ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz".split("")];
|
|
1022
|
+
/**
|
|
1023
|
+
* Detail-subtree dispatch cadence a `PipelineStepBase` step gets when its
|
|
1024
|
+
* definition literal doesn't declare one explicitly. Conservative default
|
|
1025
|
+
* for the "specialized, opt-in" classifier steps that predate the cadence
|
|
1026
|
+
* table — dispatch once per track, cap at 3 attempts.
|
|
1027
|
+
*/
|
|
1028
|
+
var DEFAULT_STEP_CADENCE = {
|
|
1029
|
+
trigger: "once",
|
|
1030
|
+
maxPerTrack: 3
|
|
1031
|
+
};
|
|
1032
|
+
var PipelineStepBase = class {
|
|
1033
|
+
definition;
|
|
1034
|
+
constructor(definition) {
|
|
1035
|
+
this.definition = {
|
|
1036
|
+
...definition,
|
|
1037
|
+
cadence: definition.cadence ?? DEFAULT_STEP_CADENCE
|
|
1038
|
+
};
|
|
1039
|
+
}
|
|
1040
|
+
getConfigSchema() {
|
|
1041
|
+
return [];
|
|
1042
|
+
}
|
|
1043
|
+
};
|
|
1044
|
+
var ObjectDetectionStep = class {
|
|
1045
|
+
definition = {
|
|
1046
|
+
id: "object-detection",
|
|
1047
|
+
name: "Object Detection",
|
|
1048
|
+
slot: "detector",
|
|
1049
|
+
postprocessor: "yolo",
|
|
1050
|
+
extractMode: "full-frame",
|
|
1051
|
+
inputClasses: null,
|
|
1052
|
+
outputClasses: [
|
|
1053
|
+
"person",
|
|
1054
|
+
"vehicle",
|
|
1055
|
+
"animal"
|
|
1056
|
+
],
|
|
1057
|
+
models: [...OBJECT_DETECTION_MODELS],
|
|
1058
|
+
defaultModelId: "yolo26n",
|
|
1059
|
+
defaultConfidence: .5,
|
|
1060
|
+
labels: COCO_80_LABELS.map((l) => l.id),
|
|
1061
|
+
classMap: COCO_TO_MACRO
|
|
1062
|
+
};
|
|
1063
|
+
getConfigSchema() {
|
|
1064
|
+
return [
|
|
1065
|
+
{
|
|
1066
|
+
type: "multiselect",
|
|
1067
|
+
key: "enabledMacroClasses",
|
|
1068
|
+
label: "Enabled Classes",
|
|
1069
|
+
description: "Which macro classes to detect (empty = all three).",
|
|
1070
|
+
options: [
|
|
1071
|
+
{
|
|
1072
|
+
value: "person",
|
|
1073
|
+
label: "Person"
|
|
1074
|
+
},
|
|
1075
|
+
{
|
|
1076
|
+
value: "vehicle",
|
|
1077
|
+
label: "Vehicle"
|
|
1078
|
+
},
|
|
1079
|
+
{
|
|
1080
|
+
value: "animal",
|
|
1081
|
+
label: "Animal"
|
|
1082
|
+
}
|
|
1083
|
+
],
|
|
1084
|
+
default: [
|
|
1085
|
+
"person",
|
|
1086
|
+
"vehicle",
|
|
1087
|
+
"animal"
|
|
1088
|
+
]
|
|
1089
|
+
},
|
|
1090
|
+
{
|
|
1091
|
+
type: "slider",
|
|
1092
|
+
key: "minConfidencePerson",
|
|
1093
|
+
label: "Min Confidence — Person",
|
|
1094
|
+
description: "Drop person predictions below this score.",
|
|
1095
|
+
min: 0,
|
|
1096
|
+
max: 1,
|
|
1097
|
+
step: .05,
|
|
1098
|
+
default: .5,
|
|
1099
|
+
showValue: true
|
|
1100
|
+
},
|
|
1101
|
+
{
|
|
1102
|
+
type: "slider",
|
|
1103
|
+
key: "minConfidenceVehicle",
|
|
1104
|
+
label: "Min Confidence — Vehicle",
|
|
1105
|
+
description: "Drop vehicle predictions below this score.",
|
|
1106
|
+
min: 0,
|
|
1107
|
+
max: 1,
|
|
1108
|
+
step: .05,
|
|
1109
|
+
default: .5,
|
|
1110
|
+
showValue: true
|
|
1111
|
+
},
|
|
1112
|
+
{
|
|
1113
|
+
type: "slider",
|
|
1114
|
+
key: "minConfidenceAnimal",
|
|
1115
|
+
label: "Min Confidence — Animal",
|
|
1116
|
+
description: "Drop animal predictions below this score.",
|
|
1117
|
+
min: 0,
|
|
1118
|
+
max: 1,
|
|
1119
|
+
step: .05,
|
|
1120
|
+
default: .5,
|
|
1121
|
+
showValue: true
|
|
1122
|
+
}
|
|
1123
|
+
];
|
|
1124
|
+
}
|
|
1125
|
+
};
|
|
1126
|
+
var AudioClassifierStep = class {
|
|
1127
|
+
definition = {
|
|
1128
|
+
id: "audio-classifier",
|
|
1129
|
+
name: "Audio Classifier",
|
|
1130
|
+
slot: "audio-classifier",
|
|
1131
|
+
postprocessor: "yamnet",
|
|
1132
|
+
extractMode: "full-frame",
|
|
1133
|
+
inputClasses: null,
|
|
1134
|
+
outputClasses: ["audio"],
|
|
1135
|
+
models: [...AUDIO_CLASSIFIER_MODELS],
|
|
1136
|
+
defaultModelId: "yamnet-onnx",
|
|
1137
|
+
defaultModelIdByFormat: { coreml: "apple-soundanalysis" },
|
|
1138
|
+
enabledByDefault: true,
|
|
1139
|
+
defaultConfidence: .3
|
|
1140
|
+
};
|
|
1141
|
+
getConfigSchema() {
|
|
1142
|
+
return [{
|
|
1143
|
+
type: "multiselect",
|
|
1144
|
+
key: "enabledAudioClasses",
|
|
1145
|
+
label: "Audio Classes",
|
|
1146
|
+
description: "Which audio macro classes to surface (empty = all).",
|
|
1147
|
+
options: AUDIO_MACRO_LABELS.map((l) => ({
|
|
1148
|
+
value: l.id,
|
|
1149
|
+
label: `${l.icon ?? ""} ${l.name}`.trim()
|
|
1150
|
+
})),
|
|
1151
|
+
default: [
|
|
1152
|
+
"scream",
|
|
1153
|
+
"crying",
|
|
1154
|
+
"dog",
|
|
1155
|
+
"alarm",
|
|
1156
|
+
"doorbell",
|
|
1157
|
+
"glass_breaking",
|
|
1158
|
+
"gunshot",
|
|
1159
|
+
"siren",
|
|
1160
|
+
"footsteps"
|
|
1161
|
+
]
|
|
1162
|
+
}, {
|
|
1163
|
+
type: "slider",
|
|
1164
|
+
key: "minConfidence",
|
|
1165
|
+
label: "Min Confidence",
|
|
1166
|
+
description: "Drop predictions whose score is below this threshold.",
|
|
1167
|
+
default: .3,
|
|
1168
|
+
min: 0,
|
|
1169
|
+
max: 1,
|
|
1170
|
+
step: .05,
|
|
1171
|
+
showValue: true
|
|
1172
|
+
}];
|
|
1173
|
+
}
|
|
1174
|
+
};
|
|
1175
|
+
var AnimalClassifierStep = class {
|
|
1176
|
+
definition = {
|
|
1177
|
+
id: "animal-classifier",
|
|
1178
|
+
name: "Animal Classifier",
|
|
1179
|
+
slot: "classifier",
|
|
1180
|
+
postprocessor: "softmax",
|
|
1181
|
+
extractMode: "crop-roi",
|
|
1182
|
+
inputClasses: ["animal"],
|
|
1183
|
+
outputClasses: ["animal-type"],
|
|
1184
|
+
models: [...ANIMAL_CLASSIFIER_MODELS],
|
|
1185
|
+
defaultModelId: "animals-10",
|
|
1186
|
+
enabledByDefault: false,
|
|
1187
|
+
defaultConfidence: .2,
|
|
1188
|
+
labels: [...ANIMAL_10_LABELS],
|
|
1189
|
+
cadence: {
|
|
1190
|
+
trigger: "once",
|
|
1191
|
+
maxPerTrack: 3,
|
|
1192
|
+
stickyOnConfidence: .7
|
|
1193
|
+
}
|
|
1194
|
+
};
|
|
1195
|
+
getConfigSchema() {
|
|
1196
|
+
return [{
|
|
1197
|
+
type: "multiselect",
|
|
1198
|
+
key: "enabledClasses",
|
|
1199
|
+
label: "Species",
|
|
1200
|
+
description: "Which animal species to keep (empty = all). The classifier still scores every label but only the selected classes are returned on the detection.",
|
|
1201
|
+
options: ANIMAL_10_LABELS.map((l) => ({
|
|
1202
|
+
value: l,
|
|
1203
|
+
label: l
|
|
1204
|
+
})),
|
|
1205
|
+
default: []
|
|
1206
|
+
}, {
|
|
1207
|
+
type: "slider",
|
|
1208
|
+
key: "minConfidence",
|
|
1209
|
+
label: "Min Confidence",
|
|
1210
|
+
description: "Drop classifications whose score is below this threshold.",
|
|
1211
|
+
default: this.definition.defaultConfidence,
|
|
1212
|
+
min: 0,
|
|
1213
|
+
max: 1,
|
|
1214
|
+
step: .05,
|
|
1215
|
+
showValue: true
|
|
1216
|
+
}];
|
|
1217
|
+
}
|
|
1218
|
+
};
|
|
1219
|
+
var ClassifierWithMinConfidence = class extends PipelineStepBase {
|
|
1220
|
+
getConfigSchema() {
|
|
1221
|
+
return [{
|
|
1222
|
+
type: "slider",
|
|
1223
|
+
key: "minConfidence",
|
|
1224
|
+
label: "Min Confidence",
|
|
1225
|
+
description: "Drop classifications whose score is below this threshold.",
|
|
1226
|
+
default: this.definition.defaultConfidence,
|
|
1227
|
+
min: 0,
|
|
1228
|
+
max: 1,
|
|
1229
|
+
step: .05,
|
|
1230
|
+
showValue: true
|
|
1231
|
+
}];
|
|
1232
|
+
}
|
|
1233
|
+
};
|
|
1234
|
+
var STEP_FACE_DETECTION = new PipelineStepBase({
|
|
1235
|
+
id: "face-detection",
|
|
1236
|
+
name: "Face Detection",
|
|
1237
|
+
slot: "cropper",
|
|
1238
|
+
postprocessor: "scrfd",
|
|
1239
|
+
extractMode: "crop-roi",
|
|
1240
|
+
inputClasses: ["person"],
|
|
1241
|
+
outputClasses: ["face"],
|
|
1242
|
+
models: [...FACE_DETECTION_MODELS],
|
|
1243
|
+
defaultModelId: "scrfd-2.5g",
|
|
1244
|
+
defaultConfidence: .5,
|
|
1245
|
+
defaultMinParentScore: .7,
|
|
1246
|
+
cadence: {
|
|
1247
|
+
trigger: "improve",
|
|
1248
|
+
minIntervalMs: 1e3
|
|
1249
|
+
}
|
|
1250
|
+
});
|
|
1251
|
+
var STEP_FACE_EMBEDDING = new PipelineStepBase({
|
|
1252
|
+
id: "face-embedding",
|
|
1253
|
+
name: "Face Embedding",
|
|
1254
|
+
slot: "classifier",
|
|
1255
|
+
postprocessor: "arcface",
|
|
1256
|
+
extractMode: "crop-roi",
|
|
1257
|
+
inputClasses: ["face"],
|
|
1258
|
+
outputClasses: ["identity"],
|
|
1259
|
+
models: [...FACE_EMBEDDING_MODELS],
|
|
1260
|
+
defaultModelId: "arcface-r100",
|
|
1261
|
+
defaultConfidence: 0,
|
|
1262
|
+
cadence: {
|
|
1263
|
+
trigger: "improve",
|
|
1264
|
+
minIntervalMs: 1e3,
|
|
1265
|
+
stickyOnConfidence: .6
|
|
1266
|
+
}
|
|
1267
|
+
});
|
|
1268
|
+
var STEP_CLIP_EMBEDDING = new PipelineStepBase({
|
|
1269
|
+
id: "clip-embedding",
|
|
1270
|
+
name: "CLIP Embedding",
|
|
1271
|
+
slot: "classifier",
|
|
1272
|
+
postprocessor: "clip",
|
|
1273
|
+
extractMode: "crop-roi",
|
|
1274
|
+
inputClasses: [
|
|
1275
|
+
"person",
|
|
1276
|
+
"vehicle",
|
|
1277
|
+
"animal"
|
|
1278
|
+
],
|
|
1279
|
+
outputClasses: ["embedding"],
|
|
1280
|
+
models: [...CLIP_EMBEDDING_MODELS],
|
|
1281
|
+
defaultModelId: "mobileclip-s1",
|
|
1282
|
+
enabledByDefault: true,
|
|
1283
|
+
backfillIntoExistingOverrides: true,
|
|
1284
|
+
defaultConfidence: 0,
|
|
1285
|
+
defaultMinParentScore: .4,
|
|
1286
|
+
group: "CLIP",
|
|
1287
|
+
cadence: { trigger: "improve" }
|
|
1288
|
+
});
|
|
1289
|
+
var STEP_PLATE_DETECTION = new PipelineStepBase({
|
|
1290
|
+
id: "plate-detection",
|
|
1291
|
+
name: "Plate Detection",
|
|
1292
|
+
slot: "cropper",
|
|
1293
|
+
postprocessor: "yolo",
|
|
1294
|
+
extractMode: "crop-roi",
|
|
1295
|
+
inputClasses: ["vehicle"],
|
|
1296
|
+
outputClasses: ["plate"],
|
|
1297
|
+
models: [...PLATE_DETECTION_MODELS],
|
|
1298
|
+
defaultModelId: "yolov8n-plate",
|
|
1299
|
+
defaultConfidence: .5,
|
|
1300
|
+
defaultMinParentScore: .7,
|
|
1301
|
+
labels: ["plate"],
|
|
1302
|
+
cadence: {
|
|
1303
|
+
trigger: "once",
|
|
1304
|
+
maxPerTrack: 5
|
|
1305
|
+
}
|
|
1306
|
+
});
|
|
1307
|
+
var STEP_PLATE_OCR = new PipelineStepBase({
|
|
1308
|
+
id: "plate-ocr",
|
|
1309
|
+
name: "Plate OCR",
|
|
1310
|
+
slot: "classifier",
|
|
1311
|
+
postprocessor: "ctc",
|
|
1312
|
+
extractMode: "crop-roi",
|
|
1313
|
+
inputClasses: ["plate"],
|
|
1314
|
+
outputClasses: ["plate-text"],
|
|
1315
|
+
models: [...PLATE_OCR_MODELS],
|
|
1316
|
+
defaultModelId: "vgg-english-g2",
|
|
1317
|
+
defaultConfidence: 0,
|
|
1318
|
+
charset: [...VGG_G2_CHARSET],
|
|
1319
|
+
cadence: {
|
|
1320
|
+
trigger: "once",
|
|
1321
|
+
maxPerTrack: 5
|
|
1322
|
+
}
|
|
1323
|
+
});
|
|
1324
|
+
var STEP_BIRD_CLASSIFIER = new ClassifierWithMinConfidence({
|
|
1325
|
+
id: "bird-classifier",
|
|
1326
|
+
name: "Bird Classifier",
|
|
1327
|
+
slot: "classifier",
|
|
1328
|
+
postprocessor: "softmax",
|
|
1329
|
+
extractMode: "crop-roi",
|
|
1330
|
+
inputClasses: ["animal"],
|
|
1331
|
+
outputClasses: ["species"],
|
|
1332
|
+
models: [...BIRD_CLASSIFIER_MODELS],
|
|
1333
|
+
defaultModelId: "bird-nabirds-404",
|
|
1334
|
+
enabledByDefault: false,
|
|
1335
|
+
defaultConfidence: .2,
|
|
1336
|
+
cadence: {
|
|
1337
|
+
trigger: "once",
|
|
1338
|
+
maxPerTrack: 3,
|
|
1339
|
+
stickyOnConfidence: .7
|
|
1340
|
+
}
|
|
1341
|
+
});
|
|
1342
|
+
var STEP_VEHICLE_CLASSIFIER = new ClassifierWithMinConfidence({
|
|
1343
|
+
id: "vehicle-classifier",
|
|
1344
|
+
name: "Vehicle Classifier",
|
|
1345
|
+
slot: "classifier",
|
|
1346
|
+
postprocessor: "softmax",
|
|
1347
|
+
extractMode: "crop-roi",
|
|
1348
|
+
inputClasses: ["vehicle"],
|
|
1349
|
+
outputClasses: ["vehicle-type"],
|
|
1350
|
+
models: [...VEHICLE_CLASSIFIER_MODELS],
|
|
1351
|
+
defaultModelId: "vehicle-type-efficientnet",
|
|
1352
|
+
enabledByDefault: false,
|
|
1353
|
+
defaultConfidence: .3,
|
|
1354
|
+
cadence: {
|
|
1355
|
+
trigger: "once",
|
|
1356
|
+
maxPerTrack: 3,
|
|
1357
|
+
stickyOnConfidence: .7
|
|
1358
|
+
}
|
|
1359
|
+
});
|
|
1360
|
+
var STEP_SEGMENTATION = new PipelineStepBase({
|
|
1361
|
+
id: "segmentation",
|
|
1362
|
+
name: "Segmentation",
|
|
1363
|
+
slot: "refiner",
|
|
1364
|
+
postprocessor: "saliency",
|
|
1365
|
+
extractMode: "crop-roi",
|
|
1366
|
+
inputClasses: null,
|
|
1367
|
+
outputClasses: ["mask"],
|
|
1368
|
+
models: [...SEGMENTATION_REFINER_MODELS],
|
|
1369
|
+
defaultModelId: "u2netp",
|
|
1370
|
+
enabledByDefault: false,
|
|
1371
|
+
defaultConfidence: 0,
|
|
1372
|
+
group: "Segmentation"
|
|
1373
|
+
});
|
|
1374
|
+
new PipelineStepBase({
|
|
1375
|
+
id: "instance-segmentation",
|
|
1376
|
+
name: "Instance Segmentation",
|
|
1377
|
+
slot: "refiner",
|
|
1378
|
+
postprocessor: "yolo-seg",
|
|
1379
|
+
extractMode: "full-frame",
|
|
1380
|
+
inputClasses: null,
|
|
1381
|
+
outputClasses: ["mask"],
|
|
1382
|
+
models: [...INSTANCE_SEGMENTATION_MODELS],
|
|
1383
|
+
defaultModelId: "yolo26n-seg",
|
|
1384
|
+
enabledByDefault: false,
|
|
1385
|
+
defaultConfidence: .25,
|
|
1386
|
+
labels: COCO_80_LABELS.map((l) => l.id),
|
|
1387
|
+
classMap: COCO_TO_MACRO,
|
|
1388
|
+
group: "Segmentation"
|
|
1389
|
+
});
|
|
1390
|
+
var STEP_OBJECT_DETECTION_INSTANCE = new ObjectDetectionStep();
|
|
1391
|
+
var STEP_AUDIO_CLASSIFIER_INSTANCE = new AudioClassifierStep();
|
|
1392
|
+
var ALL_PIPELINE_STEPS = [
|
|
1393
|
+
STEP_OBJECT_DETECTION_INSTANCE,
|
|
1394
|
+
STEP_FACE_DETECTION,
|
|
1395
|
+
STEP_FACE_EMBEDDING,
|
|
1396
|
+
STEP_CLIP_EMBEDDING,
|
|
1397
|
+
STEP_PLATE_DETECTION,
|
|
1398
|
+
STEP_PLATE_OCR,
|
|
1399
|
+
new AnimalClassifierStep(),
|
|
1400
|
+
STEP_BIRD_CLASSIFIER,
|
|
1401
|
+
STEP_VEHICLE_CLASSIFIER,
|
|
1402
|
+
STEP_SEGMENTATION,
|
|
1403
|
+
STEP_AUDIO_CLASSIFIER_INSTANCE
|
|
1404
|
+
];
|
|
1405
|
+
/** Compat: flat array of StepDefinition for existing consumers */
|
|
1406
|
+
var ALL_STEPS = ALL_PIPELINE_STEPS.map((s) => s.definition);
|
|
1407
|
+
var STEP_MAP = new Map(ALL_PIPELINE_STEPS.map((s) => [s.definition.id, s]));
|
|
1408
|
+
/**
|
|
1409
|
+
* Look up a pipeline step by ID (class instance with getConfigSchema).
|
|
1410
|
+
* @throws if the step ID is not registered.
|
|
1411
|
+
*/
|
|
1412
|
+
function getStep(stepId) {
|
|
1413
|
+
const step = STEP_MAP.get(stepId);
|
|
1414
|
+
if (!step) throw new Error(`Unknown pipeline step: "${stepId}". Available: ${ALL_PIPELINE_STEPS.map((s) => s.definition.id).join(", ")}`);
|
|
1415
|
+
return step;
|
|
1416
|
+
}
|
|
1417
|
+
/**
|
|
1418
|
+
* Look up a step definition by ID (compat shortcut).
|
|
1419
|
+
* @throws if the step ID is not registered.
|
|
1420
|
+
*/
|
|
1421
|
+
function getStepDefinition(stepId) {
|
|
1422
|
+
return getStep(stepId).definition;
|
|
1423
|
+
}
|
|
1424
|
+
/**
|
|
1425
|
+
* Get the default model ID for a step given the current model format.
|
|
1426
|
+
*
|
|
1427
|
+
* Resolution order:
|
|
1428
|
+
* 1. `def.defaultModelIdByFormat?.[format]` — the step's declared
|
|
1429
|
+
* per-format preference — if that model exists in `def.models` AND
|
|
1430
|
+
* has a build for `format`.
|
|
1431
|
+
* 2. `def.defaultModelId` — the step's plain declared default — if it
|
|
1432
|
+
* exists in `def.models` AND has a build for `format`.
|
|
1433
|
+
* 3. The smallest-by-size model among those with a `format` build
|
|
1434
|
+
* (legacy fallback, preserved for steps/formats with no declared
|
|
1435
|
+
* preference reachable).
|
|
1436
|
+
* 4. `def.defaultModelId` unchanged, when ZERO models have a `format`
|
|
1437
|
+
* build — an unloadable case flagged elsewhere, not resolved here.
|
|
1438
|
+
*/
|
|
1439
|
+
function getDefaultModelForFormat(stepId, format) {
|
|
1440
|
+
const def = getStepDefinition(stepId);
|
|
1441
|
+
const hasFormatBuild = (modelId) => def.models.find((m) => m.id === modelId)?.formats[format] !== void 0;
|
|
1442
|
+
const declaredForFormat = def.defaultModelIdByFormat?.[format];
|
|
1443
|
+
if (declaredForFormat !== void 0 && hasFormatBuild(declaredForFormat)) return declaredForFormat;
|
|
1444
|
+
if (hasFormatBuild(def.defaultModelId)) return def.defaultModelId;
|
|
1445
|
+
const available = def.models.filter((m) => m.formats[format] && m.legacy !== true);
|
|
1446
|
+
if (available.length === 0) return def.defaultModelId;
|
|
1447
|
+
return [...available].toSorted((a, b) => {
|
|
1448
|
+
return (a.formats[format]?.sizeMB ?? Infinity) - (b.formats[format]?.sizeMB ?? Infinity);
|
|
1449
|
+
})[0].id;
|
|
1450
|
+
}
|
|
1451
|
+
/**
|
|
1452
|
+
* The single node-local authority for turning a CHOSEN model (or none) into a
|
|
1453
|
+
* concrete model that has a build for this node's engine `format`. Fuses the
|
|
1454
|
+
* two previously-duplicated decisions (orchestrator `pickDefaultModelId` +
|
|
1455
|
+
* provider `substituteIncompatibleModels`):
|
|
1456
|
+
*
|
|
1457
|
+
* - chosen === null → the addon's default model for `format`.
|
|
1458
|
+
* - chosen is a catalog model → keep it if it has a `format` build; else
|
|
1459
|
+
* the addon's default model for `format`
|
|
1460
|
+
* (a substitution — the caller surfaces it).
|
|
1461
|
+
* - chosen is NOT in the catalog (custom/unknown) → returned unchanged; the
|
|
1462
|
+
* operator owns a custom model's format compat.
|
|
1463
|
+
* - unknown stepId → chosen returned unchanged (never throws).
|
|
1464
|
+
*/
|
|
1465
|
+
function resolveModelForFormat(stepId, chosenModelId, format) {
|
|
1466
|
+
let def;
|
|
1467
|
+
try {
|
|
1468
|
+
def = getStepDefinition(stepId);
|
|
1469
|
+
} catch {
|
|
1470
|
+
return chosenModelId ?? "";
|
|
1471
|
+
}
|
|
1472
|
+
if (chosenModelId === null) return getDefaultModelForFormat(stepId, format);
|
|
1473
|
+
const entry = def.models.find((m) => m.id === chosenModelId);
|
|
1474
|
+
if (!entry) return chosenModelId;
|
|
1475
|
+
if (entry.formats[format]) return chosenModelId;
|
|
1476
|
+
return getDefaultModelForFormat(stepId, format);
|
|
1477
|
+
}
|
|
1478
|
+
//#endregion
|
|
1479
|
+
export { getStepDefinition as a, getStep as i, ALL_STEPS as n, resolveModelForFormat as o, getDefaultModelForFormat as r, ALL_PIPELINE_STEPS as t };
|