oidn-web 0.3.4 → 0.4.0

package/benchmarks/results/profile-fp16-tiled-encoder.json

@@ -0,0 +1,369 @@
+{
+  "generatedAt": "2026-08-20T04:04:59.440Z",
+  "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": 8.399999998509884,
+      "timesMs": [
+        34.899999998509884,
+        35.100000001490116,
+        34.69999999552965
+      ],
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 34.69999999552965,
+        "medianMs": 34.899999998509884,
+        "meanMs": 34.899999998509884,
+        "p95Ms": 35.100000001490116,
+        "maxMs": 35.100000001490116
+      },
+      "messages": []
+    },
+    {
+      "label": "WGSL FP32",
+      "bundle": "current",
+      "baseline": false,
+      "precision": "fp32",
+      "initializationMs": 356.80000000447035,
+      "timesMs": [
+        41.099999994039536,
+        41.400000005960464,
+        41.29999999701977
+      ],
+      "executionProfile": {
+        "totalMs": 40.597770000000004,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.050539
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.956359
+          },
+          {
+            "id": "enc_conv1b",
+            "durationMs": 4.478892
+          },
+          {
+            "id": "pool1",
+            "durationMs": 0.4196
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.66644
+          },
+          {
+            "id": "enc_conv2b",
+            "durationMs": 2.496472
+          },
+          {
+            "id": "pool2",
+            "durationMs": 0.054413
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.876822
+          },
+          {
+            "id": "enc_conv3b",
+            "durationMs": 1.163387
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.018123
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.479221
+          },
+          {
+            "id": "enc_conv4b",
+            "durationMs": 0.714082
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.008416
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.292273
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.387976
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 1.462536
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.716039
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 2.723084
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 1.16043
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 5.146477
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 2.504555
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 7.877478
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 4.449561
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.494595
+          }
+        ]
+      },
+      "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": 42.20000000298023,
+          "tileTimeMs": {
+            "min": 42.20000000298023,
+            "median": 42.20000000298023,
+            "mean": 42.20000000298023,
+            "max": 42.20000000298023
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 41.099999994039536,
+        "medianMs": 41.29999999701977,
+        "meanMs": 41.266666665673256,
+        "p95Ms": 41.400000005960464,
+        "maxMs": 41.400000005960464
+      },
+      "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,
+      "timesMs": [
+        30.599999994039536,
+        30.5,
+        30.5
+      ],
+      "executionProfile": {
+        "totalMs": 29.810637,
+        "layers": [
+          {
+            "id": "input-pack",
+            "durationMs": 0.054996
+          },
+          {
+            "id": "enc_conv1a",
+            "durationMs": 0.835991
+          },
+          {
+            "id": "pool1",
+            "durationMs": 3.879471
+          },
+          {
+            "id": "enc_conv2a",
+            "durationMs": 1.450454
+          },
+          {
+            "id": "pool2",
+            "durationMs": 2.147411
+          },
+          {
+            "id": "enc_conv3a",
+            "durationMs": 0.836782
+          },
+          {
+            "id": "pool3",
+            "durationMs": 0.91011
+          },
+          {
+            "id": "enc_conv4a",
+            "durationMs": 0.419225
+          },
+          {
+            "id": "pool4",
+            "durationMs": 0.589672
+          },
+          {
+            "id": "enc_conv5a",
+            "durationMs": 0.265775
+          },
+          {
+            "id": "enc_conv5b",
+            "durationMs": 0.348396
+          },
+          {
+            "id": "dec_conv4a",
+            "durationMs": 0.947025
+          },
+          {
+            "id": "dec_conv4b",
+            "durationMs": 0.627087
+          },
+          {
+            "id": "dec_conv3a",
+            "durationMs": 1.698063
+          },
+          {
+            "id": "dec_conv3b",
+            "durationMs": 1.013688
+          },
+          {
+            "id": "dec_conv2a",
+            "durationMs": 3.015232
+          },
+          {
+            "id": "dec_conv2b",
+            "durationMs": 2.234697
+          },
+          {
+            "id": "dec_conv1a",
+            "durationMs": 4.385025
+          },
+          {
+            "id": "dec_conv1b",
+            "durationMs": 3.918302
+          },
+          {
+            "id": "dec_conv1c",
+            "durationMs": 0.233235
+          }
+        ]
+      },
+      "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": 31.5,
+          "tileTimeMs": {
+            "min": 31.400000005960464,
+            "median": 31.400000005960464,
+            "mean": 31.400000005960464,
+            "max": 31.400000005960464
+          }
+        }
+      },
+      "supportsFP16": true,
+      "adapter": {
+        "vendor": "apple",
+        "architecture": "metal-3",
+        "device": "",
+        "description": ""
+      },
+      "summary": {
+        "minMs": 30.5,
+        "medianMs": 30.5,
+        "meanMs": 30.533333331346512,
+        "p95Ms": 30.599999994039536,
+        "maxMs": 30.599999994039536
+      },
+      "messages": [],
+      "validation": {
+        "sampleCount": 6144,
+        "meanAbsoluteError": 0.001425126083631767,
+        "rootMeanSquaredError": 0.002315957223847781,
+        "maxAbsoluteError": 0.03298163414001465,
+        "passed": true,
+        "thresholds": {
+          "maxMeanError": 0.005,
+          "maxAbsoluteError": 0.05
+        }
+      }
+    }
+  ]
+}

package/benchmarks/results/profile-fp16-tiled-encoder.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) | 8.40 | 34.90 | 35.10 | 1.00x | fp32 |
+| WGSL FP32 | 356.80 | 41.30 | 41.40 | 0.85x | fp32 |
+| WGSL FP16 | 23.00 | 30.50 | 30.60 | 1.14x | 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 | 1.43e-3 | 2.32e-3 | 3.30e-2 | 6144 |
+
+## GPU hot layers
+
+### WGSL FP32
+
+Profiled GPU total: 40.60 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 7.88 |
+| dec_conv2a | 5.15 |
+| enc_conv1b | 4.48 |
+| dec_conv1b | 4.45 |
+| dec_conv3a | 2.72 |
+
+### WGSL FP16
+
+Profiled GPU total: 29.81 ms
+
+| Node | GPU ms |
+| --- | ---: |
+| dec_conv1a | 4.39 |
+| dec_conv1b | 3.92 |
+| pool1 | 3.88 |
+| dec_conv2a | 3.02 |
+| dec_conv2b | 2.23 |