oidn-web 0.3.4 → 0.4.0

package/benchmarks/results/profile-fast-init.json

@@ -0,0 +1,385 @@
+{
+  "generatedAt": "2026-08-20T04:13:39.756Z",
+  "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": 8.700000002980232,
+      "timesMs": [
+        35.29999999701977,
+        34.79999999701977,
+        34.70000000298023
+      ],
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 34.70000000298023,
+        "medianMs": 34.79999999701977,
+        "meanMs": 34.93333333233992,
+        "p95Ms": 35.29999999701977,
+        "maxMs": 35.29999999701977
+      },
+      "messages": []
+    },
+    {
+      "label": "WGSL FP32",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp32",
+      "initializationMs": 37.900000005960464,
+      "timesMs": [
+        35.899999998509884,
+        35.79999999701977,
+        35.80000000447035
+      ],
+      "executionProfile": {
+        "totalMs": 35.491794999999996,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.050704
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.839743
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.786178
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.415415
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.404142
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 2.268341
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.054578
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.766001
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.989602
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.01804
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.524902
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.611144
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.008834
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.259057
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.335089
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 1.248242
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.610268
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 2.48711
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 0.990102
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 4.369117
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 2.294795
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 6.910012
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.753974
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.496405
+          }
+        ]
+      },
+      "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": 37.19999999552965,
+          "tileTimeMs": {
+            "min": 37.19999999552965,
+            "median": 37.19999999552965,
+            "mean": 37.19999999552965,
+            "max": 37.19999999552965
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 35.79999999701977,
+        "medianMs": 35.80000000447035,
+        "meanMs": 35.833333333333336,
+        "p95Ms": 35.899999998509884,
+        "maxMs": 35.899999998509884
+      },
+      "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": 30.100000001490116,
+      "timesMs": [
+        27.700000002980232,
+        27.599999994039536,
+        27.700000002980232
+      ],
+      "executionProfile": {
+        "totalMs": 26.970954000000003,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.044871
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.618309
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 3.203075
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.339255
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.164668
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 1.999828
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.050329
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.675637
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 0.84091
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.019539
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.340588
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.530193
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.011499
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.183898
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.233393
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 0.951814
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.533193
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 1.699276
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 0.839161
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 2.978014
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 1.969207
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 4.433818
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.116417
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.194062
+          }
+        ]
+      },
+      "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": 28.5,
+          "tileTimeMs": {
+            "min": 28.5,
+            "median": 28.5,
+            "mean": 28.5,
+            "max": 28.5
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 27.599999994039536,
+        "medianMs": 27.700000002980232,
+        "meanMs": 27.666666666666668,
+        "p95Ms": 27.700000002980232,
+        "maxMs": 27.700000002980232
+      },
+      "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-fast-init.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) | 8.70 | 34.80 | 35.30 | 1.00x | fp32 |
+| WGSL FP32 | 37.90 | 35.80 | 35.90 | 0.97x | fp32 |
+| WGSL FP16 | 30.10 | 27.70 | 27.70 | 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: 35.49 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 6.91 |
+| dec_conv2a | 4.37 |
+| enc_conv1b | 3.79 |
+| dec_conv1b | 3.75 |
+| dec_conv3a | 2.49 |
+
+### WGSL FP16
+
+Profiled GPU total: 26.97 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 4.43 |
+| enc_conv1b | 3.20 |
+| dec_conv1b | 3.12 |
+| dec_conv2a | 2.98 |
+| enc_conv2b | 2.00 |