oidn-web 0.4.0 → 0.5.0

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 (106) hide show
  1. package/CHANGELOG.md +45 -0
  2. package/README.md +83 -19
  3. package/dist/oidn.js +3239 -2642
  4. package/dist/oidn.umd.cjs +512 -296
  5. package/lib/UNet.d.ts +54 -20
  6. package/lib/UNet.js +194 -118
  7. package/lib/UNet.js.map +1 -1
  8. package/lib/backend.d.ts +1 -8
  9. package/lib/backend.js +1 -9
  10. package/lib/backend.js.map +1 -1
  11. package/lib/finalRgbShader.d.ts +13 -0
  12. package/lib/finalRgbShader.js +160 -0
  13. package/lib/finalRgbShader.js.map +1 -0
  14. package/lib/graphOptimizer.js +1 -2
  15. package/lib/graphOptimizer.js.map +1 -1
  16. package/lib/hdrTransfer.d.ts +14 -0
  17. package/lib/hdrTransfer.js +61 -0
  18. package/lib/hdrTransfer.js.map +1 -0
  19. package/lib/main.d.ts +16 -7
  20. package/lib/main.js +4 -5
  21. package/lib/main.js.map +1 -1
  22. package/lib/nativeUNet.d.ts +39 -3
  23. package/lib/nativeUNet.js +449 -120
  24. package/lib/nativeUNet.js.map +1 -1
  25. package/lib/process.d.ts +5 -11
  26. package/lib/process.js +35 -49
  27. package/lib/process.js.map +1 -1
  28. package/lib/tileScheduler.d.ts +32 -4
  29. package/lib/tileScheduler.js +133 -20
  30. package/lib/tileScheduler.js.map +1 -1
  31. package/package.json +9 -2
  32. package/src/UNet.ts +287 -158
  33. package/src/backend.ts +1 -14
  34. package/src/finalRgbShader.ts +186 -0
  35. package/src/graphOptimizer.ts +1 -2
  36. package/src/hdrTransfer.ts +88 -0
  37. package/src/main.ts +28 -13
  38. package/src/nativeUNet.ts +515 -116
  39. package/src/process.ts +43 -70
  40. package/src/tileScheduler.ts +216 -24
  41. package/benchmarks/compare.mjs +0 -651
  42. package/benchmarks/leak.mjs +0 -255
  43. package/benchmarks/results/before-spatial.json +0 -391
  44. package/benchmarks/results/before-spatial.md +0 -47
  45. package/benchmarks/results/int8-scan.json +0 -2007
  46. package/benchmarks/results/int8-scan.md +0 -160
  47. package/benchmarks/results/int8-w8a8-scan.json +0 -2007
  48. package/benchmarks/results/int8-w8a8-scan.md +0 -160
  49. package/benchmarks/results/int8-weight-channel.json +0 -1413
  50. package/benchmarks/results/int8-weight-channel.md +0 -118
  51. package/benchmarks/results/int8-weight-only.json +0 -1437
  52. package/benchmarks/results/int8-weight-only.md +0 -118
  53. package/benchmarks/results/kernel-webnn-final.json +0 -1115
  54. package/benchmarks/results/kernel-webnn-final.md +0 -104
  55. package/benchmarks/results/latest-optimized.json +0 -375
  56. package/benchmarks/results/latest-optimized.md +0 -47
  57. package/benchmarks/results/latest.json +0 -391
  58. package/benchmarks/results/latest.md +0 -47
  59. package/benchmarks/results/profile-baseline.json +0 -331
  60. package/benchmarks/results/profile-baseline.md +0 -12
  61. package/benchmarks/results/profile-conv2x.json +0 -331
  62. package/benchmarks/results/profile-conv2x.md +0 -12
  63. package/benchmarks/results/profile-fast-init.json +0 -385
  64. package/benchmarks/results/profile-fast-init.md +0 -47
  65. package/benchmarks/results/profile-fp16-fma.json +0 -369
  66. package/benchmarks/results/profile-fp16-fma.md +0 -47
  67. package/benchmarks/results/profile-fp16-tiled-decoder.json +0 -369
  68. package/benchmarks/results/profile-fp16-tiled-decoder.md +0 -47
  69. package/benchmarks/results/profile-fp16-tiled-encoder.json +0 -369
  70. package/benchmarks/results/profile-fp16-tiled-encoder.md +0 -47
  71. package/benchmarks/results/profile-fp16-unfused-pool.json +0 -385
  72. package/benchmarks/results/profile-fp16-unfused-pool.md +0 -47
  73. package/benchmarks/results/profile-input-major.json +0 -347
  74. package/benchmarks/results/profile-input-major.md +0 -12
  75. package/benchmarks/results/profile-k16.json +0 -347
  76. package/benchmarks/results/profile-k16.md +0 -12
  77. package/benchmarks/results/profile-k4.json +0 -347
  78. package/benchmarks/results/profile-k4.md +0 -12
  79. package/benchmarks/results/profile-pool-reuse.json +0 -331
  80. package/benchmarks/results/profile-pool-reuse.md +0 -12
  81. package/benchmarks/results/profile-precompiled.json +0 -385
  82. package/benchmarks/results/profile-precompiled.md +0 -47
  83. package/benchmarks/results/profile-static-channels.json +0 -385
  84. package/benchmarks/results/profile-static-channels.md +0 -47
  85. package/benchmarks/results/profile-static-io.json +0 -385
  86. package/benchmarks/results/profile-static-io.md +0 -47
  87. package/benchmarks/results/profile-tiled-conv.json +0 -331
  88. package/benchmarks/results/profile-tiled-conv.md +0 -12
  89. package/benchmarks/results/profile-tiled-decoder.json +0 -347
  90. package/benchmarks/results/profile-tiled-decoder.md +0 -12
  91. package/benchmarks/results/profile-tiled-matmul.json +0 -331
  92. package/benchmarks/results/profile-tiled-matmul.md +0 -12
  93. package/benchmarks/results/profile-unfused-decoder.json +0 -379
  94. package/benchmarks/results/profile-unfused-decoder.md +0 -12
  95. package/benchmarks/results/profile-unfused-pool.json +0 -347
  96. package/benchmarks/results/profile-unfused-pool.md +0 -12
  97. package/benchmarks/results/spatial-auto.json +0 -575
  98. package/benchmarks/results/spatial-auto.md +0 -61
  99. package/benchmarks/results/subgroup-smoke.json +0 -1094
  100. package/benchmarks/results/subgroup-smoke.md +0 -104
  101. package/benchmarks/results/webnn-smoke.json +0 -739
  102. package/benchmarks/results/webnn-smoke.md +0 -76
  103. package/scripts/inspect-model.mjs +0 -64
  104. package/tests/modelSpec.test.mjs +0 -128
  105. package/tests/resourceLifecycle.test.mjs +0 -383
  106. package/tests/tileScheduler.test.mjs +0 -90
@@ -1,369 +0,0 @@
1
- {
2
- "generatedAt": "2026-08-20T04:04:59.440Z",
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": 8.399999998509884,
27
- "timesMs": [
28
- 34.899999998509884,
29
- 35.100000001490116,
30
- 34.69999999552965
31
- ],
32
- "supportsFP16": true,
33
- "adapter": {
34
- "vendor": "apple",
35
- "architecture": "metal-3",
36
- "device": "",
37
- "description": ""
38
- },
39
- "summary": {
40
- "minMs": 34.69999999552965,
41
- "medianMs": 34.899999998509884,
42
- "meanMs": 34.899999998509884,
43
- "p95Ms": 35.100000001490116,
44
- "maxMs": 35.100000001490116
45
- },
46
- "messages": []
47
- },
48
- {
49
- "label": "WGSL FP32",
50
- "bundle": "current",
51
- "baseline": false,
52
- "precision": "fp32",
53
- "initializationMs": 356.80000000447035,
54
- "timesMs": [
55
- 41.099999994039536,
56
- 41.400000005960464,
57
- 41.29999999701977
58
- ],
59
- "executionProfile": {
60
- "totalMs": 40.597770000000004,
61
- "layers": [
62
- {
63
- "id": "input-pack",
64
- "durationMs": 0.050539
65
- },
66
- {
67
- "id": "enc_conv1a",
68
- "durationMs": 0.956359
69
- },
70
- {
71
- "id": "enc_conv1b",
72
- "durationMs": 4.478892
73
- },
74
- {
75
- "id": "pool1",
76
- "durationMs": 0.4196
77
- },
78
- {
79
- "id": "enc_conv2a",
80
- "durationMs": 1.66644
81
- },
82
- {
83
- "id": "enc_conv2b",
84
- "durationMs": 2.496472
85
- },
86
- {
87
- "id": "pool2",
88
- "durationMs": 0.054413
89
- },
90
- {
91
- "id": "enc_conv3a",
92
- "durationMs": 0.876822
93
- },
94
- {
95
- "id": "enc_conv3b",
96
- "durationMs": 1.163387
97
- },
98
- {
99
- "id": "pool3",
100
- "durationMs": 0.018123
101
- },
102
- {
103
- "id": "enc_conv4a",
104
- "durationMs": 0.479221
105
- },
106
- {
107
- "id": "enc_conv4b",
108
- "durationMs": 0.714082
109
- },
110
- {
111
- "id": "pool4",
112
- "durationMs": 0.008416
113
- },
114
- {
115
- "id": "enc_conv5a",
116
- "durationMs": 0.292273
117
- },
118
- {
119
- "id": "enc_conv5b",
120
- "durationMs": 0.387976
121
- },
122
- {
123
- "id": "dec_conv4a",
124
- "durationMs": 1.462536
125
- },
126
- {
127
- "id": "dec_conv4b",
128
- "durationMs": 0.716039
129
- },
130
- {
131
- "id": "dec_conv3a",
132
- "durationMs": 2.723084
133
- },
134
- {
135
- "id": "dec_conv3b",
136
- "durationMs": 1.16043
137
- },
138
- {
139
- "id": "dec_conv2a",
140
- "durationMs": 5.146477
141
- },
142
- {
143
- "id": "dec_conv2b",
144
- "durationMs": 2.504555
145
- },
146
- {
147
- "id": "dec_conv1a",
148
- "durationMs": 7.877478
149
- },
150
- {
151
- "id": "dec_conv1b",
152
- "durationMs": 4.449561
153
- },
154
- {
155
- "id": "dec_conv1c",
156
- "durationMs": 0.494595
157
- }
158
- ]
159
- },
160
- "runtimeInfo": {
161
- "configuredEngine": "wgsl",
162
- "gpuEngine": "wgsl",
163
- "precision": "fp32",
164
- "model": "oidn-unet-large-v1",
165
- "modelFamily": "oidn-unet-large",
166
- "inputChannels": 9,
167
- "dynamicTile": {
168
- "enabled": false,
169
- "currentTileSize": 512,
170
- "minTileSize": 256,
171
- "maxTileSize": 512,
172
- "targetTileTimeMs": 16
173
- },
174
- "lastExecution": {
175
- "width": 512,
176
- "height": 512,
177
- "tileWidth": 512,
178
- "tileHeight": 512,
179
- "tileCount": 1,
180
- "durationMs": 42.20000000298023,
181
- "tileTimeMs": {
182
- "min": 42.20000000298023,
183
- "median": 42.20000000298023,
184
- "mean": 42.20000000298023,
185
- "max": 42.20000000298023
186
- }
187
- }
188
- },
189
- "supportsFP16": true,
190
- "adapter": {
191
- "vendor": "apple",
192
- "architecture": "metal-3",
193
- "device": "",
194
- "description": ""
195
- },
196
- "summary": {
197
- "minMs": 41.099999994039536,
198
- "medianMs": 41.29999999701977,
199
- "meanMs": 41.266666665673256,
200
- "p95Ms": 41.400000005960464,
201
- "maxMs": 41.400000005960464
202
- },
203
- "messages": [],
204
- "validation": {
205
- "sampleCount": 6144,
206
- "meanAbsoluteError": 0.0000011096847932397698,
207
- "rootMeanSquaredError": 0.0000018071588418628667,
208
- "maxAbsoluteError": 0.000013589859008789062,
209
- "passed": true,
210
- "thresholds": {
211
- "maxMeanError": 0.0001,
212
- "maxAbsoluteError": 0.001
213
- }
214
- }
215
- },
216
- {
217
- "label": "WGSL FP16",
218
- "bundle": "current",
219
- "baseline": false,
220
- "precision": "fp16",
221
- "initializationMs": 23,
222
- "timesMs": [
223
- 30.599999994039536,
224
- 30.5,
225
- 30.5
226
- ],
227
- "executionProfile": {
228
- "totalMs": 29.810637,
229
- "layers": [
230
- {
231
- "id": "input-pack",
232
- "durationMs": 0.054996
233
- },
234
- {
235
- "id": "enc_conv1a",
236
- "durationMs": 0.835991
237
- },
238
- {
239
- "id": "pool1",
240
- "durationMs": 3.879471
241
- },
242
- {
243
- "id": "enc_conv2a",
244
- "durationMs": 1.450454
245
- },
246
- {
247
- "id": "pool2",
248
- "durationMs": 2.147411
249
- },
250
- {
251
- "id": "enc_conv3a",
252
- "durationMs": 0.836782
253
- },
254
- {
255
- "id": "pool3",
256
- "durationMs": 0.91011
257
- },
258
- {
259
- "id": "enc_conv4a",
260
- "durationMs": 0.419225
261
- },
262
- {
263
- "id": "pool4",
264
- "durationMs": 0.589672
265
- },
266
- {
267
- "id": "enc_conv5a",
268
- "durationMs": 0.265775
269
- },
270
- {
271
- "id": "enc_conv5b",
272
- "durationMs": 0.348396
273
- },
274
- {
275
- "id": "dec_conv4a",
276
- "durationMs": 0.947025
277
- },
278
- {
279
- "id": "dec_conv4b",
280
- "durationMs": 0.627087
281
- },
282
- {
283
- "id": "dec_conv3a",
284
- "durationMs": 1.698063
285
- },
286
- {
287
- "id": "dec_conv3b",
288
- "durationMs": 1.013688
289
- },
290
- {
291
- "id": "dec_conv2a",
292
- "durationMs": 3.015232
293
- },
294
- {
295
- "id": "dec_conv2b",
296
- "durationMs": 2.234697
297
- },
298
- {
299
- "id": "dec_conv1a",
300
- "durationMs": 4.385025
301
- },
302
- {
303
- "id": "dec_conv1b",
304
- "durationMs": 3.918302
305
- },
306
- {
307
- "id": "dec_conv1c",
308
- "durationMs": 0.233235
309
- }
310
- ]
311
- },
312
- "runtimeInfo": {
313
- "configuredEngine": "wgsl",
314
- "gpuEngine": "wgsl",
315
- "precision": "fp16",
316
- "model": "oidn-unet-large-v1",
317
- "modelFamily": "oidn-unet-large",
318
- "inputChannels": 9,
319
- "dynamicTile": {
320
- "enabled": false,
321
- "currentTileSize": 512,
322
- "minTileSize": 256,
323
- "maxTileSize": 512,
324
- "targetTileTimeMs": 16
325
- },
326
- "lastExecution": {
327
- "width": 512,
328
- "height": 512,
329
- "tileWidth": 512,
330
- "tileHeight": 512,
331
- "tileCount": 1,
332
- "durationMs": 31.5,
333
- "tileTimeMs": {
334
- "min": 31.400000005960464,
335
- "median": 31.400000005960464,
336
- "mean": 31.400000005960464,
337
- "max": 31.400000005960464
338
- }
339
- }
340
- },
341
- "supportsFP16": true,
342
- "adapter": {
343
- "vendor": "apple",
344
- "architecture": "metal-3",
345
- "device": "",
346
- "description": ""
347
- },
348
- "summary": {
349
- "minMs": 30.5,
350
- "medianMs": 30.5,
351
- "meanMs": 30.533333331346512,
352
- "p95Ms": 30.599999994039536,
353
- "maxMs": 30.599999994039536
354
- },
355
- "messages": [],
356
- "validation": {
357
- "sampleCount": 6144,
358
- "meanAbsoluteError": 0.001425126083631767,
359
- "rootMeanSquaredError": 0.002315957223847781,
360
- "maxAbsoluteError": 0.03298163414001465,
361
- "passed": true,
362
- "thresholds": {
363
- "maxMeanError": 0.005,
364
- "maxAbsoluteError": 0.05
365
- }
366
- }
367
- }
368
- ]
369
- }
@@ -1,47 +0,0 @@
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) | 8.40 | 34.90 | 35.10 | 1.00x | fp32 |
11
- | WGSL FP32 | 356.80 | 41.30 | 41.40 | 0.85x | fp32 |
12
- | WGSL FP16 | 23.00 | 30.50 | 30.60 | 1.14x | fp16 |
13
-
14
- ## Output validation
15
-
16
- Compared against sampled TFJS FP32 output.
17
-
18
- | Runtime | Status | MAE | RMSE | Max error | Samples |
19
- | --- | --- | ---: | ---: | ---: | ---: |
20
- | WGSL FP32 | pass | 1.11e-6 | 1.81e-6 | 1.36e-5 | 6144 |
21
- | WGSL FP16 | pass | 1.43e-3 | 2.32e-3 | 3.30e-2 | 6144 |
22
-
23
- ## GPU hot layers
24
-
25
- ### WGSL FP32
26
-
27
- Profiled GPU total: 40.60 ms
28
-
29
- | Node | GPU ms |
30
- | --- | ---: |
31
- | dec_conv1a | 7.88 |
32
- | dec_conv2a | 5.15 |
33
- | enc_conv1b | 4.48 |
34
- | dec_conv1b | 4.45 |
35
- | dec_conv3a | 2.72 |
36
-
37
- ### WGSL FP16
38
-
39
- Profiled GPU total: 29.81 ms
40
-
41
- | Node | GPU ms |
42
- | --- | ---: |
43
- | dec_conv1a | 4.39 |
44
- | dec_conv1b | 3.92 |
45
- | pool1 | 3.88 |
46
- | dec_conv2a | 3.02 |
47
- | dec_conv2b | 2.23 |