oidn-web 0.3.4 → 0.4.0

package/benchmarks/results/spatial-auto.json

@@ -0,0 +1,575 @@
+{
+  "generatedAt": "2026-08-20T04:48:54.112Z",
+  "currentCommit": "e6b8577b1c69cc8ce979022717984aec5b9ea881",
+  "baselineCommit": "136c00c34aa5cf4561b4137034dbd643177f0a5f",
+  "settings": {
+    "width": 512,
+    "height": 512,
+    "tileSize": 512,
+    "warmup": 2,
+    "runs": 5,
+    "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,
+      "timesMs": [
+        37.5,
+        37,
+        37.100000001490116,
+        37.29999999701977,
+        37.70000000298023
+      ],
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 37,
+        "medianMs": 37.29999999701977,
+        "meanMs": 37.32000000029802,
+        "p95Ms": 37.70000000298023,
+        "maxMs": 37.70000000298023
+      },
+      "messages": []
+    },
+    {
+      "label": "WGSL FP32",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp32",
+      "initializationMs": 38.600000001490116,
+      "timesMs": [
+        37.900000005960464,
+        38,
+        37.29999999701977,
+        38.100000001490116,
+        37.29999999701977
+      ],
+      "executionProfile": {
+        "totalMs": 37.46062,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.051206
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 1.406145
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.828992
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.417481
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.402811
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 2.168693
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.39194
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.943749
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.996538
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.018374
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.399774
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.609222
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.008582
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.256989
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.337401
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 1.243484
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.609264
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 2.330977
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 1.270983
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 4.547001
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 2.158151
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 7.394786
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 4.171225
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.496852
+          }
+        ]
+      },
+      "runtimeInfo": {
+        "configuredEngine": "wgsl",
+        "gpuEngine": "wgsl",
+        "precision": "fp32",
+        "kernel": {
+          "configured": "auto",
+          "maxSpatialInputBlocks": 0
+        },
+        "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": 39.20000000298023,
+          "tileTimeMs": {
+            "min": 39.20000000298023,
+            "median": 39.20000000298023,
+            "mean": 39.20000000298023,
+            "max": 39.20000000298023
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 37.29999999701977,
+        "medianMs": 37.900000005960464,
+        "meanMs": 37.72000000029802,
+        "p95Ms": 38.100000001490116,
+        "maxMs": 38.100000001490116
+      },
+      "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 Direct",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp16",
+      "kernel": "direct",
+      "initializationMs": 30.69999999552965,
+      "timesMs": [
+        30.299999997019768,
+        30.30000000447035,
+        29.399999998509884,
+        30.100000001490116,
+        29.899999998509884
+      ],
+      "executionProfile": {
+        "totalMs": 29.140974999999997,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.446681
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.634003
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.123184
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.337396
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.162887
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 1.97238
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.050289
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 1.250132
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.896529
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.018166
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.342479
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.531717
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.010958
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.187323
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.23082
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 0.945734
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.535134
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 1.69773
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 0.835783
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 3.644611
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 1.975421
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 5.004736
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.114102
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.19278
+          }
+        ]
+      },
+      "runtimeInfo": {
+        "configuredEngine": "wgsl",
+        "gpuEngine": "wgsl",
+        "precision": "fp16",
+        "kernel": {
+          "configured": "direct",
+          "maxSpatialInputBlocks": 40
+        },
+        "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": 30.69999999552965,
+          "tileTimeMs": {
+            "min": 30.69999999552965,
+            "median": 30.69999999552965,
+            "mean": 30.69999999552965,
+            "max": 30.69999999552965
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 29.399999998509884,
+        "medianMs": 30.100000001490116,
+        "meanMs": 30,
+        "p95Ms": 30.30000000447035,
+        "maxMs": 30.30000000447035
+      },
+      "messages": [],
+      "validation": {
+        "sampleCount": 6144,
+        "meanAbsoluteError": 0.0010210334021394374,
+        "rootMeanSquaredError": 0.001609680379854656,
+        "maxAbsoluteError": 0.012395620346069336,
+        "passed": true,
+        "thresholds": {
+          "maxMeanError": 0.005,
+          "maxAbsoluteError": 0.05
+        }
+      }
+    },
+    {
+      "label": "WGSL FP16 Adaptive",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp16",
+      "kernel": "auto",
+      "initializationMs": 123.89999999850988,
+      "timesMs": [
+        65.89999999850988,
+        65.60000000149012,
+        64.5,
+        65.5,
+        65.20000000298023
+      ],
+      "executionProfile": {
+        "totalMs": 65.040055,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.044998
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.720644
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 6.022776
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.34803
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 2.047354
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 4.902892
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.054665
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 1.430082
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 2.580421
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.015666
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 1.410749
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.933222
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.053623
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.191161
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.237993
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 0.96847
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.535859
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 1.717239
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 2.881037
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 3.642306
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 4.314618
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 23.615654
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 6.005524
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.365072
+          }
+        ]
+      },
+      "runtimeInfo": {
+        "configuredEngine": "wgsl",
+        "gpuEngine": "wgsl",
+        "precision": "fp16",
+        "kernel": {
+          "configured": "auto",
+          "maxSpatialInputBlocks": 40
+        },
+        "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": 66.60000000149012,
+          "tileTimeMs": {
+            "min": 66.60000000149012,
+            "median": 66.60000000149012,
+            "mean": 66.60000000149012,
+            "max": 66.60000000149012
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 64.5,
+        "medianMs": 65.5,
+        "meanMs": 65.34000000059605,
+        "p95Ms": 65.89999999850988,
+        "maxMs": 65.89999999850988
+      },
+      "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/spatial-auto.md

@@ -0,0 +1,61 @@
+# oidn-web benchmark
+
+- Current: `e6b8577b1c69cc8ce979022717984aec5b9ea881`
+- TFJS baseline: `136c00c34aa5cf4561b4137034dbd643177f0a5f`
+- Input: 512x512, fixed tile 512, 5 runs after 2 warmup(s)
+- Adapter: apple
+
+| Runtime | Init ms | Median ms | P95 ms | Speedup | Precision |
+| --- | ---: | ---: | ---: | ---: | --- |
+| TFJS (136c00c) | 9.00 | 37.30 | 37.70 | 1.00x | fp32 |
+| WGSL FP32 | 38.60 | 37.90 | 38.10 | 0.98x | fp32 |
+| WGSL FP16 Direct | 30.70 | 30.10 | 30.30 | 1.24x | fp16 |
+| WGSL FP16 Adaptive | 123.90 | 65.50 | 65.90 | 0.57x | 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 Direct | pass | 1.02e-3 | 1.61e-3 | 1.24e-2 | 6144 |
+| WGSL FP16 Adaptive | pass | 1.02e-3 | 1.61e-3 | 1.24e-2 | 6144 |
+
+## GPU hot layers
+
+### WGSL FP32
+
+Profiled GPU total: 37.46 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 7.39 |
+| dec_conv2a | 4.55 |
+| dec_conv1b | 4.17 |
+| enc_conv1b | 3.83 |
+| dec_conv3a | 2.33 |
+
+### WGSL FP16 Direct
+
+Profiled GPU total: 29.14 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 5.00 |
+| dec_conv2a | 3.64 |
+| enc_conv1b | 3.12 |
+| dec_conv1b | 3.11 |
+| dec_conv2b | 1.98 |
+
+### WGSL FP16 Adaptive
+
+Profiled GPU total: 65.04 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 23.62 |
+| enc_conv1b | 6.02 |
+| dec_conv1b | 6.01 |
+| enc_conv2b | 4.90 |
+| dec_conv2b | 4.31 |