oidn-web 0.3.4 → 0.4.0

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.
Files changed (122) hide show
  1. package/CHANGELOG.md +49 -0
  2. package/README.md +140 -4
  3. package/benchmarks/compare.mjs +651 -0
  4. package/benchmarks/leak.mjs +255 -0
  5. package/benchmarks/results/before-spatial.json +391 -0
  6. package/benchmarks/results/before-spatial.md +47 -0
  7. package/benchmarks/results/int8-scan.json +2007 -0
  8. package/benchmarks/results/int8-scan.md +160 -0
  9. package/benchmarks/results/int8-w8a8-scan.json +2007 -0
  10. package/benchmarks/results/int8-w8a8-scan.md +160 -0
  11. package/benchmarks/results/int8-weight-channel.json +1413 -0
  12. package/benchmarks/results/int8-weight-channel.md +118 -0
  13. package/benchmarks/results/int8-weight-only.json +1437 -0
  14. package/benchmarks/results/int8-weight-only.md +118 -0
  15. package/benchmarks/results/kernel-webnn-final.json +1115 -0
  16. package/benchmarks/results/kernel-webnn-final.md +104 -0
  17. package/benchmarks/results/latest-optimized.json +375 -0
  18. package/benchmarks/results/latest-optimized.md +47 -0
  19. package/benchmarks/results/latest.json +391 -0
  20. package/benchmarks/results/latest.md +47 -0
  21. package/benchmarks/results/profile-baseline.json +331 -0
  22. package/benchmarks/results/profile-baseline.md +12 -0
  23. package/benchmarks/results/profile-conv2x.json +331 -0
  24. package/benchmarks/results/profile-conv2x.md +12 -0
  25. package/benchmarks/results/profile-fast-init.json +385 -0
  26. package/benchmarks/results/profile-fast-init.md +47 -0
  27. package/benchmarks/results/profile-fp16-fma.json +369 -0
  28. package/benchmarks/results/profile-fp16-fma.md +47 -0
  29. package/benchmarks/results/profile-fp16-tiled-decoder.json +369 -0
  30. package/benchmarks/results/profile-fp16-tiled-decoder.md +47 -0
  31. package/benchmarks/results/profile-fp16-tiled-encoder.json +369 -0
  32. package/benchmarks/results/profile-fp16-tiled-encoder.md +47 -0
  33. package/benchmarks/results/profile-fp16-unfused-pool.json +385 -0
  34. package/benchmarks/results/profile-fp16-unfused-pool.md +47 -0
  35. package/benchmarks/results/profile-input-major.json +347 -0
  36. package/benchmarks/results/profile-input-major.md +12 -0
  37. package/benchmarks/results/profile-k16.json +347 -0
  38. package/benchmarks/results/profile-k16.md +12 -0
  39. package/benchmarks/results/profile-k4.json +347 -0
  40. package/benchmarks/results/profile-k4.md +12 -0
  41. package/benchmarks/results/profile-pool-reuse.json +331 -0
  42. package/benchmarks/results/profile-pool-reuse.md +12 -0
  43. package/benchmarks/results/profile-precompiled.json +385 -0
  44. package/benchmarks/results/profile-precompiled.md +47 -0
  45. package/benchmarks/results/profile-static-channels.json +385 -0
  46. package/benchmarks/results/profile-static-channels.md +47 -0
  47. package/benchmarks/results/profile-static-io.json +385 -0
  48. package/benchmarks/results/profile-static-io.md +47 -0
  49. package/benchmarks/results/profile-tiled-conv.json +331 -0
  50. package/benchmarks/results/profile-tiled-conv.md +12 -0
  51. package/benchmarks/results/profile-tiled-decoder.json +347 -0
  52. package/benchmarks/results/profile-tiled-decoder.md +12 -0
  53. package/benchmarks/results/profile-tiled-matmul.json +331 -0
  54. package/benchmarks/results/profile-tiled-matmul.md +12 -0
  55. package/benchmarks/results/profile-unfused-decoder.json +379 -0
  56. package/benchmarks/results/profile-unfused-decoder.md +12 -0
  57. package/benchmarks/results/profile-unfused-pool.json +347 -0
  58. package/benchmarks/results/profile-unfused-pool.md +12 -0
  59. package/benchmarks/results/spatial-auto.json +575 -0
  60. package/benchmarks/results/spatial-auto.md +61 -0
  61. package/benchmarks/results/subgroup-smoke.json +1094 -0
  62. package/benchmarks/results/subgroup-smoke.md +104 -0
  63. package/benchmarks/results/webnn-smoke.json +739 -0
  64. package/benchmarks/results/webnn-smoke.md +76 -0
  65. package/dist/oidn.js +4189 -22603
  66. package/dist/oidn.umd.cjs +784 -5807
  67. package/lib/UNet.d.ts +66 -15
  68. package/lib/UNet.js +162 -257
  69. package/lib/UNet.js.map +1 -1
  70. package/lib/WGPUComputePass.d.ts +1 -1
  71. package/lib/WGPUComputePass.js +6 -4
  72. package/lib/WGPUComputePass.js.map +1 -1
  73. package/lib/backend.d.ts +8 -4
  74. package/lib/backend.js +36 -44
  75. package/lib/backend.js.map +1 -1
  76. package/lib/graphOptimizer.d.ts +54 -0
  77. package/lib/graphOptimizer.js +216 -0
  78. package/lib/graphOptimizer.js.map +1 -0
  79. package/lib/main.d.ts +33 -10
  80. package/lib/main.js +5 -0
  81. package/lib/main.js.map +1 -1
  82. package/lib/modelSpec.d.ts +80 -0
  83. package/lib/modelSpec.js +270 -0
  84. package/lib/modelSpec.js.map +1 -0
  85. package/lib/nativeUNet.d.ts +67 -0
  86. package/lib/nativeUNet.js +1735 -0
  87. package/lib/nativeUNet.js.map +1 -0
  88. package/lib/process.js +38 -35
  89. package/lib/process.js.map +1 -1
  90. package/lib/resourceTracker.d.ts +26 -0
  91. package/lib/resourceTracker.js +65 -0
  92. package/lib/resourceTracker.js.map +1 -0
  93. package/lib/tileScheduler.d.ts +33 -0
  94. package/lib/tileScheduler.js +86 -0
  95. package/lib/tileScheduler.js.map +1 -0
  96. package/lib/webnnUNet.d.ts +52 -0
  97. package/lib/webnnUNet.js +535 -0
  98. package/lib/webnnUNet.js.map +1 -0
  99. package/package.json +9 -5
  100. package/scripts/inspect-model.mjs +64 -0
  101. package/src/UNet.ts +236 -339
  102. package/src/WGPUComputePass.ts +6 -4
  103. package/src/backend.ts +42 -55
  104. package/src/graphOptimizer.ts +301 -0
  105. package/src/main.ts +71 -11
  106. package/src/modelSpec.ts +414 -0
  107. package/src/nativeUNet.ts +2256 -0
  108. package/src/process.ts +38 -36
  109. package/src/resourceTracker.ts +94 -0
  110. package/src/tileScheduler.ts +138 -0
  111. package/src/webnnUNet.ts +812 -0
  112. package/tests/modelSpec.test.mjs +128 -0
  113. package/tests/resourceLifecycle.test.mjs +383 -0
  114. package/tests/tileScheduler.test.mjs +90 -0
  115. package/lib/helper.d.ts +0 -4
  116. package/lib/helper.js +0 -33
  117. package/lib/helper.js.map +0 -1
  118. package/lib/kernels.d.ts +0 -1
  119. package/lib/kernels.js +0 -26
  120. package/lib/kernels.js.map +0 -1
  121. package/src/helper.ts +0 -43
  122. package/src/kernels.ts +0 -31
@@ -0,0 +1,379 @@
1
+ {
2
+ "generatedAt": "2026-08-20T03:55:26.300Z",
3
+ "currentCommit": "5d9c1c8771e2a03c90bfadafec4b5dd28fada446",
4
+ "baselineCommit": "136c00c34aa5cf4561b4137034dbd643177f0a5f",
5
+ "settings": {
6
+ "width": 512,
7
+ "height": 512,
8
+ "tileSize": 512,
9
+ "warmup": 1,
10
+ "runs": 3,
11
+ "model": "rt_hdr_calb_cnrm_large.tza",
12
+ "gpuQueueCompletionIncluded": true
13
+ },
14
+ "browser": "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome",
15
+ "adapter": {
16
+ "vendor": "apple",
17
+ "architecture": "metal-3",
18
+ "device": "",
19
+ "description": ""
20
+ },
21
+ "results": [
22
+ {
23
+ "label": "TFJS (136c00c)",
24
+ "bundle": "baseline",
25
+ "baseline": true,
26
+ "initializationMs": 8.299999997019768,
27
+ "timesMs": [
28
+ 34.79999999701977,
29
+ 34.80000000447035,
30
+ 35.099999994039536
31
+ ],
32
+ "supportsFP16": true,
33
+ "adapter": {
34
+ "vendor": "apple",
35
+ "architecture": "metal-3",
36
+ "device": "",
37
+ "description": ""
38
+ },
39
+ "summary": {
40
+ "minMs": 34.79999999701977,
41
+ "medianMs": 34.80000000447035,
42
+ "meanMs": 34.899999998509884,
43
+ "p95Ms": 35.099999994039536,
44
+ "maxMs": 35.099999994039536
45
+ },
46
+ "messages": []
47
+ },
48
+ {
49
+ "label": "WGSL FP32",
50
+ "bundle": "current",
51
+ "baseline": false,
52
+ "precision": "fp32",
53
+ "initializationMs": 364.1000000014901,
54
+ "timesMs": [
55
+ 44.399999998509884,
56
+ 44.29999999701977,
57
+ 44.30000000447035
58
+ ],
59
+ "executionProfile": {
60
+ "totalMs": 43.114608999999994,
61
+ "layers": [
62
+ {
63
+ "id": "input-pack",
64
+ "durationMs": 0.051331
65
+ },
66
+ {
67
+ "id": "enc_conv1a",
68
+ "durationMs": 1.075535
69
+ },
70
+ {
71
+ "id": "enc_conv1b",
72
+ "durationMs": 4.384257
73
+ },
74
+ {
75
+ "id": "pool1",
76
+ "durationMs": 0.427856
77
+ },
78
+ {
79
+ "id": "enc_conv2a",
80
+ "durationMs": 1.691547
81
+ },
82
+ {
83
+ "id": "enc_conv2b",
84
+ "durationMs": 2.616881
85
+ },
86
+ {
87
+ "id": "pool2",
88
+ "durationMs": 0.058622
89
+ },
90
+ {
91
+ "id": "enc_conv3a",
92
+ "durationMs": 0.879335
93
+ },
94
+ {
95
+ "id": "enc_conv3b",
96
+ "durationMs": 1.164155
97
+ },
98
+ {
99
+ "id": "pool3",
100
+ "durationMs": 0.018041
101
+ },
102
+ {
103
+ "id": "enc_conv4a",
104
+ "durationMs": 0.479519
105
+ },
106
+ {
107
+ "id": "enc_conv4b",
108
+ "durationMs": 0.718842
109
+ },
110
+ {
111
+ "id": "pool4",
112
+ "durationMs": 0.008333
113
+ },
114
+ {
115
+ "id": "enc_conv5a",
116
+ "durationMs": 0.290444
117
+ },
118
+ {
119
+ "id": "enc_conv5b",
120
+ "durationMs": 0.399815
121
+ },
122
+ {
123
+ "id": "up4",
124
+ "durationMs": 0.012708
125
+ },
126
+ {
127
+ "id": "concat4",
128
+ "durationMs": 0.028624
129
+ },
130
+ {
131
+ "id": "dec_conv4a",
132
+ "durationMs": 1.42856
133
+ },
134
+ {
135
+ "id": "dec_conv4b",
136
+ "durationMs": 0.71805
137
+ },
138
+ {
139
+ "id": "up3",
140
+ "durationMs": 0.027998
141
+ },
142
+ {
143
+ "id": "concat3",
144
+ "durationMs": 0.077538
145
+ },
146
+ {
147
+ "id": "dec_conv3a",
148
+ "durationMs": 2.706668
149
+ },
150
+ {
151
+ "id": "dec_conv3b",
152
+ "durationMs": 1.165947
153
+ },
154
+ {
155
+ "id": "up2",
156
+ "durationMs": 0.105162
157
+ },
158
+ {
159
+ "id": "concat2",
160
+ "durationMs": 0.323193
161
+ },
162
+ {
163
+ "id": "dec_conv2a",
164
+ "durationMs": 5.015021
165
+ },
166
+ {
167
+ "id": "dec_conv2b",
168
+ "durationMs": 2.499719
169
+ },
170
+ {
171
+ "id": "up1",
172
+ "durationMs": 0.605389
173
+ },
174
+ {
175
+ "id": "concat1",
176
+ "durationMs": 1.633925
177
+ },
178
+ {
179
+ "id": "dec_conv1a",
180
+ "durationMs": 7.587153
181
+ },
182
+ {
183
+ "id": "dec_conv1b",
184
+ "durationMs": 4.422256
185
+ },
186
+ {
187
+ "id": "dec_conv1c",
188
+ "durationMs": 0.492185
189
+ }
190
+ ]
191
+ },
192
+ "runtimeInfo": {
193
+ "configuredEngine": "wgsl",
194
+ "gpuEngine": "wgsl",
195
+ "precision": "fp32",
196
+ "model": "oidn-unet-large-v1",
197
+ "modelFamily": "oidn-unet-large",
198
+ "inputChannels": 9,
199
+ "dynamicTile": {
200
+ "enabled": false,
201
+ "currentTileSize": 512,
202
+ "minTileSize": 256,
203
+ "maxTileSize": 512,
204
+ "targetTileTimeMs": 16
205
+ },
206
+ "lastExecution": {
207
+ "width": 512,
208
+ "height": 512,
209
+ "tileWidth": 512,
210
+ "tileHeight": 512,
211
+ "tileCount": 1,
212
+ "durationMs": 44.899999998509884,
213
+ "tileTimeMs": {
214
+ "min": 44.899999998509884,
215
+ "median": 44.899999998509884,
216
+ "mean": 44.899999998509884,
217
+ "max": 44.899999998509884
218
+ }
219
+ }
220
+ },
221
+ "supportsFP16": true,
222
+ "adapter": {
223
+ "vendor": "apple",
224
+ "architecture": "metal-3",
225
+ "device": "",
226
+ "description": ""
227
+ },
228
+ "summary": {
229
+ "minMs": 44.29999999701977,
230
+ "medianMs": 44.30000000447035,
231
+ "meanMs": 44.333333333333336,
232
+ "p95Ms": 44.399999998509884,
233
+ "maxMs": 44.399999998509884
234
+ },
235
+ "messages": []
236
+ },
237
+ {
238
+ "label": "WGSL FP16",
239
+ "bundle": "current",
240
+ "baseline": false,
241
+ "precision": "fp16",
242
+ "initializationMs": 24.400000005960464,
243
+ "timesMs": [
244
+ 32.20000000298023,
245
+ 32,
246
+ 31.700000002980232
247
+ ],
248
+ "executionProfile": {
249
+ "totalMs": 30.610730000000004,
250
+ "layers": [
251
+ {
252
+ "id": "input-pack",
253
+ "durationMs": 0.044912
254
+ },
255
+ {
256
+ "id": "enc_conv1a",
257
+ "durationMs": 0.736504
258
+ },
259
+ {
260
+ "id": "pool1",
261
+ "durationMs": 4.125226
262
+ },
263
+ {
264
+ "id": "enc_conv2a",
265
+ "durationMs": 1.511755
266
+ },
267
+ {
268
+ "id": "pool2",
269
+ "durationMs": 2.272299
270
+ },
271
+ {
272
+ "id": "enc_conv3a",
273
+ "durationMs": 0.756544
274
+ },
275
+ {
276
+ "id": "pool3",
277
+ "durationMs": 0.952523
278
+ },
279
+ {
280
+ "id": "enc_conv4a",
281
+ "durationMs": 0.358379
282
+ },
283
+ {
284
+ "id": "pool4",
285
+ "durationMs": 0.594353
286
+ },
287
+ {
288
+ "id": "enc_conv5a",
289
+ "durationMs": 0.195479
290
+ },
291
+ {
292
+ "id": "enc_conv5b",
293
+ "durationMs": 0.254514
294
+ },
295
+ {
296
+ "id": "dec_conv4a",
297
+ "durationMs": 0.958315
298
+ },
299
+ {
300
+ "id": "dec_conv4b",
301
+ "durationMs": 0.53736
302
+ },
303
+ {
304
+ "id": "dec_conv3a",
305
+ "durationMs": 1.762187
306
+ },
307
+ {
308
+ "id": "dec_conv3b",
309
+ "durationMs": 0.986186
310
+ },
311
+ {
312
+ "id": "dec_conv2a",
313
+ "durationMs": 3.202408
314
+ },
315
+ {
316
+ "id": "dec_conv2b",
317
+ "durationMs": 2.344416
318
+ },
319
+ {
320
+ "id": "dec_conv1a",
321
+ "durationMs": 4.699998
322
+ },
323
+ {
324
+ "id": "dec_conv1b",
325
+ "durationMs": 4.069482
326
+ },
327
+ {
328
+ "id": "dec_conv1c",
329
+ "durationMs": 0.24789
330
+ }
331
+ ]
332
+ },
333
+ "runtimeInfo": {
334
+ "configuredEngine": "wgsl",
335
+ "gpuEngine": "wgsl",
336
+ "precision": "fp16",
337
+ "model": "oidn-unet-large-v1",
338
+ "modelFamily": "oidn-unet-large",
339
+ "inputChannels": 9,
340
+ "dynamicTile": {
341
+ "enabled": false,
342
+ "currentTileSize": 512,
343
+ "minTileSize": 256,
344
+ "maxTileSize": 512,
345
+ "targetTileTimeMs": 16
346
+ },
347
+ "lastExecution": {
348
+ "width": 512,
349
+ "height": 512,
350
+ "tileWidth": 512,
351
+ "tileHeight": 512,
352
+ "tileCount": 1,
353
+ "durationMs": 32.19999999552965,
354
+ "tileTimeMs": {
355
+ "min": 32.19999999552965,
356
+ "median": 32.19999999552965,
357
+ "mean": 32.19999999552965,
358
+ "max": 32.19999999552965
359
+ }
360
+ }
361
+ },
362
+ "supportsFP16": true,
363
+ "adapter": {
364
+ "vendor": "apple",
365
+ "architecture": "metal-3",
366
+ "device": "",
367
+ "description": ""
368
+ },
369
+ "summary": {
370
+ "minMs": 31.700000002980232,
371
+ "medianMs": 32,
372
+ "meanMs": 31.966666668653488,
373
+ "p95Ms": 32.20000000298023,
374
+ "maxMs": 32.20000000298023
375
+ },
376
+ "messages": []
377
+ }
378
+ ]
379
+ }
@@ -0,0 +1,12 @@
1
+ # oidn-web benchmark
2
+
3
+ - Current: `5d9c1c8771e2a03c90bfadafec4b5dd28fada446`
4
+ - TFJS baseline: `136c00c34aa5cf4561b4137034dbd643177f0a5f`
5
+ - Input: 512x512, fixed tile 512, 3 runs after 1 warmup(s)
6
+ - Adapter: apple
7
+
8
+ | Runtime | Init ms | Median ms | P95 ms | Speedup | Precision |
9
+ | --- | ---: | ---: | ---: | ---: | --- |
10
+ | TFJS (136c00c) | 8.30 | 34.80 | 35.10 | 1.00x | fp32 |
11
+ | WGSL FP32 | 364.10 | 44.30 | 44.40 | 0.79x | fp32 |
12
+ | WGSL FP16 | 24.40 | 32.00 | 32.20 | 1.09x | fp16 |