oidn-web 0.3.4 → 0.4.0

package/benchmarks/results/profile-static-io.json

@@ -0,0 +1,385 @@
+{
+  "generatedAt": "2026-08-20T04:09:47.409Z",
+  "currentCommit": "5d9c1c8771e2a03c90bfadafec4b5dd28fada446",
+  "baselineCommit": "136c00c34aa5cf4561b4137034dbd643177f0a5f",
+  "settings": {
+    "width": 512,
+    "height": 512,
+    "tileSize": 512,
+    "warmup": 1,
+    "runs": 3,
+    "model": "rt_hdr_calb_cnrm_large.tza",
+    "gpuQueueCompletionIncluded": true
+  },
+  "browser": "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome",
+  "adapter": {
+    "vendor": "apple",
+    "architecture": "metal-3",
+    "device": "",
+    "description": ""
+  },
+  "results": [
+    {
+      "label": "TFJS (136c00c)",
+      "bundle": "baseline",
+      "baseline": true,
+      "initializationMs": 9.100000001490116,
+      "timesMs": [
+        34.79999999701977,
+        34.899999998509884,
+        34.80000000447035
+      ],
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 34.79999999701977,
+        "medianMs": 34.80000000447035,
+        "meanMs": 34.833333333333336,
+        "p95Ms": 34.899999998509884,
+        "maxMs": 34.899999998509884
+      },
+      "messages": []
+    },
+    {
+      "label": "WGSL FP32",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp32",
+      "initializationMs": 354.3999999985099,
+      "timesMs": [
+        36.100000001490116,
+        35.79999999701977,
+        35.80000000447035
+      ],
+      "executionProfile": {
+        "totalMs": 35.21059700000001,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.051164
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.838628
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.831159
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.411522
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.401186
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 2.240147
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.054913
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.771757
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.988955
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.017832
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.399856
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.610347
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.011542
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.257988
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.340025
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 1.244735
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.610265
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 2.369433
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 0.987289
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 4.370217
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 2.157109
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 6.941183
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.809076
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.494269
+          }
+        ]
+      },
+      "runtimeInfo": {
+        "configuredEngine": "wgsl",
+        "gpuEngine": "wgsl",
+        "precision": "fp32",
+        "model": "oidn-unet-large-v1",
+        "modelFamily": "oidn-unet-large",
+        "inputChannels": 9,
+        "dynamicTile": {
+          "enabled": false,
+          "currentTileSize": 512,
+          "minTileSize": 256,
+          "maxTileSize": 512,
+          "targetTileTimeMs": 16
+        },
+        "lastExecution": {
+          "width": 512,
+          "height": 512,
+          "tileWidth": 512,
+          "tileHeight": 512,
+          "tileCount": 1,
+          "durationMs": 36.899999998509884,
+          "tileTimeMs": {
+            "min": 36.899999998509884,
+            "median": 36.899999998509884,
+            "mean": 36.899999998509884,
+            "max": 36.899999998509884
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 35.79999999701977,
+        "medianMs": 35.80000000447035,
+        "meanMs": 35.90000000099341,
+        "p95Ms": 36.100000001490116,
+        "maxMs": 36.100000001490116
+      },
+      "messages": [],
+      "validation": {
+        "sampleCount": 6144,
+        "meanAbsoluteError": 0.0000011096847932397698,
+        "rootMeanSquaredError": 0.0000018071588418628667,
+        "maxAbsoluteError": 0.000013589859008789062,
+        "passed": true,
+        "thresholds": {
+          "maxMeanError": 0.0001,
+          "maxAbsoluteError": 0.001
+        }
+      }
+    },
+    {
+      "label": "WGSL FP16",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp16",
+      "initializationMs": 23.399999998509884,
+      "timesMs": [
+        27.899999998509884,
+        27.700000002980232,
+        27.69999999552965
+      ],
+      "executionProfile": {
+        "totalMs": 27.053387999999998,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.045747
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.617088
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.108226
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.340812
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.19301
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 1.97388
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.052455
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.679334
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.891821
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.017706
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.343812
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.528134
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.013957
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.184739
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.234819
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 0.956941
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.535468
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 1.712687
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 0.838949
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 3.020898
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 1.96588
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 4.427439
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.176181
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.193405
+          }
+        ]
+      },
+      "runtimeInfo": {
+        "configuredEngine": "wgsl",
+        "gpuEngine": "wgsl",
+        "precision": "fp16",
+        "model": "oidn-unet-large-v1",
+        "modelFamily": "oidn-unet-large",
+        "inputChannels": 9,
+        "dynamicTile": {
+          "enabled": false,
+          "currentTileSize": 512,
+          "minTileSize": 256,
+          "maxTileSize": 512,
+          "targetTileTimeMs": 16
+        },
+        "lastExecution": {
+          "width": 512,
+          "height": 512,
+          "tileWidth": 512,
+          "tileHeight": 512,
+          "tileCount": 1,
+          "durationMs": 28.600000001490116,
+          "tileTimeMs": {
+            "min": 28.600000001490116,
+            "median": 28.600000001490116,
+            "mean": 28.600000001490116,
+            "max": 28.600000001490116
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 27.69999999552965,
+        "medianMs": 27.700000002980232,
+        "meanMs": 27.766666665673256,
+        "p95Ms": 27.899999998509884,
+        "maxMs": 27.899999998509884
+      },
+      "messages": [],
+      "validation": {
+        "sampleCount": 6144,
+        "meanAbsoluteError": 0.0010210334021394374,
+        "rootMeanSquaredError": 0.001609680379854656,
+        "maxAbsoluteError": 0.012395620346069336,
+        "passed": true,
+        "thresholds": {
+          "maxMeanError": 0.005,
+          "maxAbsoluteError": 0.05
+        }
+      }
+    }
+  ]
+}

package/benchmarks/results/profile-static-io.md

@@ -0,0 +1,47 @@
+# oidn-web benchmark
+
+- Current: `5d9c1c8771e2a03c90bfadafec4b5dd28fada446`
+- TFJS baseline: `136c00c34aa5cf4561b4137034dbd643177f0a5f`
+- Input: 512x512, fixed tile 512, 3 runs after 1 warmup(s)
+- Adapter: apple
+
+| Runtime | Init ms | Median ms | P95 ms | Speedup | Precision |
+| --- | ---: | ---: | ---: | ---: | --- |
+| TFJS (136c00c) | 9.10 | 34.80 | 34.90 | 1.00x | fp32 |
+| WGSL FP32 | 354.40 | 35.80 | 36.10 | 0.97x | fp32 |
+| WGSL FP16 | 23.40 | 27.70 | 27.90 | 1.26x | fp16 |
+
+## Output validation
+
+Compared against sampled TFJS FP32 output.
+
+| Runtime | Status | MAE | RMSE | Max error | Samples |
+| --- | --- | ---: | ---: | ---: | ---: |
+| WGSL FP32 | pass | 1.11e-6 | 1.81e-6 | 1.36e-5 | 6144 |
+| WGSL FP16 | pass | 1.02e-3 | 1.61e-3 | 1.24e-2 | 6144 |
+
+## GPU hot layers
+
+### WGSL FP32
+
+Profiled GPU total: 35.21 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 6.94 |
+| dec_conv2a | 4.37 |
+| enc_conv1b | 3.83 |
+| dec_conv1b | 3.81 |
+| dec_conv3a | 2.37 |
+
+### WGSL FP16
+
+Profiled GPU total: 27.05 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 4.43 |
+| dec_conv1b | 3.18 |
+| enc_conv1b | 3.11 |
+| dec_conv2a | 3.02 |
+| enc_conv2b | 1.97 |