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 +1,331 @@
1
+ {
2
+ "generatedAt": "2026-08-20T03:47:38.702Z",
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": 8.5,
27
+ "timesMs": [
28
+ 37.100000001490116,
29
+ 37.29999999701977,
30
+ 37.20000000298023
31
+ ],
32
+ "supportsFP16": true,
33
+ "adapter": {
34
+ "vendor": "apple",
35
+ "architecture": "metal-3",
36
+ "device": "",
37
+ "description": ""
38
+ },
39
+ "summary": {
40
+ "minMs": 37.100000001490116,
41
+ "medianMs": 37.20000000298023,
42
+ "meanMs": 37.20000000049671,
43
+ "p95Ms": 37.29999999701977,
44
+ "maxMs": 37.29999999701977
45
+ },
46
+ "messages": []
47
+ },
48
+ {
49
+ "label": "WGSL FP32",
50
+ "bundle": "current",
51
+ "baseline": false,
52
+ "precision": "fp32",
53
+ "initializationMs": 354.20000000298023,
54
+ "timesMs": [
55
+ 61.5,
56
+ 60.5,
57
+ 60.79999999701977
58
+ ],
59
+ "executionProfile": {
60
+ "totalMs": 58.839125,
61
+ "layers": [
62
+ {
63
+ "id": "input-pack",
64
+ "durationMs": 0.059416
65
+ },
66
+ {
67
+ "id": "enc_conv1a",
68
+ "durationMs": 0.970928
69
+ },
70
+ {
71
+ "id": "pool1",
72
+ "durationMs": 12.519868
73
+ },
74
+ {
75
+ "id": "enc_conv2a",
76
+ "durationMs": 2.063562
77
+ },
78
+ {
79
+ "id": "pool2",
80
+ "durationMs": 4.253995
81
+ },
82
+ {
83
+ "id": "enc_conv3a",
84
+ "durationMs": 0.922764
85
+ },
86
+ {
87
+ "id": "pool3",
88
+ "durationMs": 1.614701
89
+ },
90
+ {
91
+ "id": "enc_conv4a",
92
+ "durationMs": 0.488943
93
+ },
94
+ {
95
+ "id": "pool4",
96
+ "durationMs": 0.916472
97
+ },
98
+ {
99
+ "id": "enc_conv5a",
100
+ "durationMs": 0.296741
101
+ },
102
+ {
103
+ "id": "enc_conv5b",
104
+ "durationMs": 0.393113
105
+ },
106
+ {
107
+ "id": "dec_conv4a",
108
+ "durationMs": 1.656659
109
+ },
110
+ {
111
+ "id": "dec_conv4b",
112
+ "durationMs": 0.730185
113
+ },
114
+ {
115
+ "id": "dec_conv3a",
116
+ "durationMs": 3.595224
117
+ },
118
+ {
119
+ "id": "dec_conv3b",
120
+ "durationMs": 1.393
121
+ },
122
+ {
123
+ "id": "dec_conv2a",
124
+ "durationMs": 7.05241
125
+ },
126
+ {
127
+ "id": "dec_conv2b",
128
+ "durationMs": 2.558214
129
+ },
130
+ {
131
+ "id": "dec_conv1a",
132
+ "durationMs": 10.421141
133
+ },
134
+ {
135
+ "id": "dec_conv1b",
136
+ "durationMs": 4.526071
137
+ },
138
+ {
139
+ "id": "dec_conv1c",
140
+ "durationMs": 2.405718
141
+ }
142
+ ]
143
+ },
144
+ "runtimeInfo": {
145
+ "configuredEngine": "wgsl",
146
+ "gpuEngine": "wgsl",
147
+ "precision": "fp32",
148
+ "model": "oidn-unet-large-v1",
149
+ "modelFamily": "oidn-unet-large",
150
+ "inputChannels": 9,
151
+ "dynamicTile": {
152
+ "enabled": false,
153
+ "currentTileSize": 512,
154
+ "minTileSize": 256,
155
+ "maxTileSize": 512,
156
+ "targetTileTimeMs": 16
157
+ },
158
+ "lastExecution": {
159
+ "width": 512,
160
+ "height": 512,
161
+ "tileWidth": 512,
162
+ "tileHeight": 512,
163
+ "tileCount": 1,
164
+ "durationMs": 60.80000000447035,
165
+ "tileTimeMs": {
166
+ "min": 60.80000000447035,
167
+ "median": 60.80000000447035,
168
+ "mean": 60.80000000447035,
169
+ "max": 60.80000000447035
170
+ }
171
+ }
172
+ },
173
+ "supportsFP16": true,
174
+ "adapter": {
175
+ "vendor": "apple",
176
+ "architecture": "metal-3",
177
+ "device": "",
178
+ "description": ""
179
+ },
180
+ "summary": {
181
+ "minMs": 60.5,
182
+ "medianMs": 60.79999999701977,
183
+ "meanMs": 60.93333333233992,
184
+ "p95Ms": 61.5,
185
+ "maxMs": 61.5
186
+ },
187
+ "messages": []
188
+ },
189
+ {
190
+ "label": "WGSL FP16",
191
+ "bundle": "current",
192
+ "baseline": false,
193
+ "precision": "fp16",
194
+ "initializationMs": 22.900000005960464,
195
+ "timesMs": [
196
+ 33.79999999701977,
197
+ 34,
198
+ 33.70000000298023
199
+ ],
200
+ "executionProfile": {
201
+ "totalMs": 33.194250000000004,
202
+ "layers": [
203
+ {
204
+ "id": "input-pack",
205
+ "durationMs": 0.055453
206
+ },
207
+ {
208
+ "id": "enc_conv1a",
209
+ "durationMs": 0.737213
210
+ },
211
+ {
212
+ "id": "pool1",
213
+ "durationMs": 4.20001
214
+ },
215
+ {
216
+ "id": "enc_conv2a",
217
+ "durationMs": 1.512506
218
+ },
219
+ {
220
+ "id": "pool2",
221
+ "durationMs": 2.816324
222
+ },
223
+ {
224
+ "id": "enc_conv3a",
225
+ "durationMs": 0.761461
226
+ },
227
+ {
228
+ "id": "pool3",
229
+ "durationMs": 0.997769
230
+ },
231
+ {
232
+ "id": "enc_conv4a",
233
+ "durationMs": 0.366003
234
+ },
235
+ {
236
+ "id": "pool4",
237
+ "durationMs": 0.592395
238
+ },
239
+ {
240
+ "id": "enc_conv5a",
241
+ "durationMs": 0.195229
242
+ },
243
+ {
244
+ "id": "enc_conv5b",
245
+ "durationMs": 0.25264
246
+ },
247
+ {
248
+ "id": "dec_conv4a",
249
+ "durationMs": 0.963606
250
+ },
251
+ {
252
+ "id": "dec_conv4b",
253
+ "durationMs": 0.537943
254
+ },
255
+ {
256
+ "id": "dec_conv3a",
257
+ "durationMs": 1.99337
258
+ },
259
+ {
260
+ "id": "dec_conv3b",
261
+ "durationMs": 1.321067
262
+ },
263
+ {
264
+ "id": "dec_conv2a",
265
+ "durationMs": 3.246278
266
+ },
267
+ {
268
+ "id": "dec_conv2b",
269
+ "durationMs": 2.345708
270
+ },
271
+ {
272
+ "id": "dec_conv1a",
273
+ "durationMs": 5.355553
274
+ },
275
+ {
276
+ "id": "dec_conv1b",
277
+ "durationMs": 4.692624
278
+ },
279
+ {
280
+ "id": "dec_conv1c",
281
+ "durationMs": 0.251098
282
+ }
283
+ ]
284
+ },
285
+ "runtimeInfo": {
286
+ "configuredEngine": "wgsl",
287
+ "gpuEngine": "wgsl",
288
+ "precision": "fp16",
289
+ "model": "oidn-unet-large-v1",
290
+ "modelFamily": "oidn-unet-large",
291
+ "inputChannels": 9,
292
+ "dynamicTile": {
293
+ "enabled": false,
294
+ "currentTileSize": 512,
295
+ "minTileSize": 256,
296
+ "maxTileSize": 512,
297
+ "targetTileTimeMs": 16
298
+ },
299
+ "lastExecution": {
300
+ "width": 512,
301
+ "height": 512,
302
+ "tileWidth": 512,
303
+ "tileHeight": 512,
304
+ "tileCount": 1,
305
+ "durationMs": 35.099999994039536,
306
+ "tileTimeMs": {
307
+ "min": 35.099999994039536,
308
+ "median": 35.099999994039536,
309
+ "mean": 35.099999994039536,
310
+ "max": 35.099999994039536
311
+ }
312
+ }
313
+ },
314
+ "supportsFP16": true,
315
+ "adapter": {
316
+ "vendor": "apple",
317
+ "architecture": "metal-3",
318
+ "device": "",
319
+ "description": ""
320
+ },
321
+ "summary": {
322
+ "minMs": 33.70000000298023,
323
+ "medianMs": 33.79999999701977,
324
+ "meanMs": 33.833333333333336,
325
+ "p95Ms": 34,
326
+ "maxMs": 34
327
+ },
328
+ "messages": []
329
+ }
330
+ ]
331
+ }
@@ -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) | 8.50 | 37.20 | 37.30 | 1.00x | fp32 |
11
+ | WGSL FP32 | 354.20 | 60.80 | 61.50 | 0.61x | fp32 |
12
+ | WGSL FP16 | 22.90 | 33.80 | 34.00 | 1.10x | fp16 |