Add device-validated Vulkan FP16 and FP32 variants

.gitattributes

@@ -37,3 +37,5 @@ coreml/u2netp_coreml_fp16.pte filter=lfs diff=lfs merge=lfs -text
 coreml/u2netp_coreml_fp32.pte filter=lfs diff=lfs merge=lfs -text
 xnnpack/u2netp_xnnpack_fp32.pte filter=lfs diff=lfs merge=lfs -text
 xnnpack/u2netp_xnnpack_int8.pte filter=lfs diff=lfs merge=lfs -text
+vulkan/u2netp_vulkan_fp16.pte filter=lfs diff=lfs merge=lfs -text
+vulkan/u2netp_vulkan_fp32.pte filter=lfs diff=lfs merge=lfs -text

README.md

@@ -25,6 +25,8 @@ revision `ac7e1c817ecab7c7dff5ce6b1abba61cd213ff29`.
 | `coreml/u2netp_coreml_fp32.pte`   | coreml  | fp32      |
 | `xnnpack/u2netp_xnnpack_fp32.pte` | xnnpack | fp32      |
 | `xnnpack/u2netp_xnnpack_int8.pte` | xnnpack | selective int8 / fp32 |
+| `vulkan/u2netp_vulkan_fp16.pte` | vulkan | fp16 with one fp32 convolution |
+| `vulkan/u2netp_vulkan_fp32.pte` | vulkan | fp32 |
 
 The exports take RGB float32 `[1,3,320,320]` input in `[0,1]` and return one
 float32 `[1,1,320,320]` foreground logit map. Image-maximum scaling and ImageNet
@@ -50,6 +52,14 @@ Full-network INT8 remains unpublished because it loses substantial foreground
 detail. See [XNNPACK precision notes](xnnpack/NOTES.md) for the selective recipe,
 calibration, validation and rejected alternatives.
 
+Vulkan FP16 preserves input normalization and one sensitive convolution in FP32.
+On a Galaxy S20+ (Android 13, Mali), native forward medians were 337 ms for this
+variant and 403 ms for Vulkan FP32, versus 316 ms for XNNPACK FP32 and 189 ms
+for selective INT8. Vulkan is optional; Android retains its XNNPACK FP32
+default. Across 13 device-tested images, Vulkan FP16 averaged 0.9945 mask IoU
+against FP32 (minimum 0.9854). See [Vulkan precision notes](vulkan/NOTES.md)
+for the export recipe, phone validation and timing limitations.
+
 ## Repository structure
 
 ```
@@ -57,8 +67,12 @@ config.json                                    28 B
 coreml/config.json                           1.5 kB
 coreml/u2netp_coreml_fp16.pte                3.3 MB
 coreml/u2netp_coreml_fp32.pte                5.5 MB
+vulkan/config.json                           1.5 kB
+vulkan/NOTES.md                              4.2 kB
+vulkan/u2netp_vulkan_fp16.pte                2.4 MB
+vulkan/u2netp_vulkan_fp32.pte                4.6 MB
 xnnpack/config.json                          1.5 kB
-xnnpack/NOTES.md                             6.5 kB
+xnnpack/NOTES.md                             6.6 kB
 xnnpack/u2netp_xnnpack_fp32.pte              4.6 MB
 xnnpack/u2netp_xnnpack_int8.pte              4.0 MB
 LICENSE                                     11.4 kB

vulkan/NOTES.md

@@ -0,0 +1,91 @@
+# Vulkan precision and Android validation
+
+Both variants accept RGB float32 `[1,3,320,320]` in `[0,1]` and return float32
+`[1,1,320,320]` foreground logits. Input normalization stays in FP32. Each
+export contains one Vulkan delegate for the network and one portable input
+`amax` reduction. Texture limits are capped at 2048 in each dimension.
+
+## FP16 with a sensitive FP32 convolution
+
+The FP16 export fuses Conv/BatchNorm pairs before casting the network to half.
+`stage2d.rebnconv1d.conv_s1` retains FP32 weights, bias and computation, with
+casts around it inside the Vulkan delegate. Global `force_fp16` is deliberately
+not used: it would also downcast this convolution and input normalization.
+
+On the connected Galaxy S20+, an entirely FP16 export changed the bicycle mask
+to 0.9418 IoU against XNNPACK FP32. Keeping this convolution in FP32 recovered
+0.9932 IoU. The entirely FP16 candidate is not published.
+
+## Device checks
+
+Validation used Samsung Galaxy S20+ SM-G986B, Android 13, Mali GPU, arm64-v8a,
+and the ExecuTorch 1.4.1 runtime/backend libraries shipped with RNE. A native
+runner linked these exact libraries, loaded each actual `.pte`, checked float32
+output type and element count, rejected non-finite output, and saved logits.
+Exported method metadata was also checked against the config.
+
+The comparison set is the same 13 images used for selective XNNPACK INT8
+validation: `bike.jpg`, `boat.jpg`, `whisk.png`, plus COCO val2017 image IDs 139,
+285, 632, 724, 776, 785, 802, 872, 885 and 1000. All-black and all-white inputs
+were also checked. These are agreement measurements against the shipped
+XNNPACK FP32 `.pte`, not ground-truth segmentation accuracy.
+
+| Variant | Mean mask IoU (13 images) | Minimum mask IoU | File size |
+| --- | --- | --- | --- |
+| Vulkan FP16 with one FP32 convolution | 0.9945 | 0.9854 | 2,437,516 B |
+| Vulkan FP32 | 0.999999 | 0.999986 | 4,646,540 B |
+
+The FP16 probability MAE was at most 0.00444 across those 13 images. Both
+variants produced finite outputs on all 15 inputs.
+
+Warm native forward measurements on the same bicycle input used three warm-up
+runs followed by ten measured runs per variant:
+
+| Variant | Median |
+| --- | --- |
+| XNNPACK FP32 | 315.6 ms |
+| XNNPACK selective INT8 | 189.1 ms |
+| Vulkan FP16 with one FP32 convolution | 337.2 ms |
+| Vulkan FP32 | 403.1 ms |
+
+These were sequential host-driven device checks, not interleaved benchmarks;
+they exclude model loading, image preprocessing and rendering. Vulkan does not
+win on this older Mali GPU. Android retains its existing XNNPACK FP32 default;
+select INT8 for the faster CPU option. Performance on other GPUs must be
+measured separately. A trial with only four half-precision convolutions was
+also slower (421.9 ms) and is not published.
+
+## Reproducing the export environment
+
+The Vulkan AOT implementation is pinned to software-mansion-labs/executorch
+`@ms/separate-backends-1.4.1`, commit
+`0de5ce9470e4a18d6d45585f69519267bbe218b9`. Torch is 2.14.0. The nested
+`pyproject.toml` records the fork dependency; see the repository Vulkan skill
+for the source-built environment used by other Vulkan models.
+
+For AOT-only export, a compatible ExecuTorch 1.4.1 wheel can provide the common
+serialization binaries while the fork supplies its Vulkan Python package.
+No host GPU runtime is needed; Android validation still uses the shipped native
+Vulkan library. Clone the fork into a directory named `executorch`, check out
+the commit above, then from export-scripts run:
+
+```bash
+uv run --project exporter_models/u2netp python - /path/to/executorch <<'PY'
+import runpy
+import sys
+from pathlib import Path
+import executorch.backends
+
+fork = Path(sys.argv[1]).resolve()
+executorch.backends.__path__.insert(0, str(fork / "backends"))
+sys.argv = [
+    "exporter", "export", "u2netp", "--backend", "vulkan", "--all-variants",
+    "--output-dir", "exports/u2netp/vulkan",
+]
+runpy.run_module("exporter.cli", run_name="__main__")
+PY
+```
+
+The release exports were byte-identical to the device-validated candidates.
+The Python wheel's native runtime cannot execute Vulkan on this Mac; it was
+not used as a substitute for phone validation.

vulkan/config.json

@@ -0,0 +1,74 @@
+{
+  "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
+  "model": "u2netp",
+  "family": "u2net",
+  "capabilities": [
+    "semantic-segmentation"
+  ],
+  "backend": "vulkan",
+  "license": "apache-2.0",
+  "variants": [
+    {
+      "file": "u2netp_vulkan_fp16.pte",
+      "precision": "fp16",
+      "methods": {
+        "forward": {
+          "inputs": [
+            {
+              "shape": [
+                1,
+                3,
+                320,
+                320
+              ],
+              "dtype": "float32"
+            }
+          ],
+          "outputs": [
+            {
+              "shape": [
+                1,
+                1,
+                320,
+                320
+              ],
+              "dtype": "float32"
+            }
+          ]
+        }
+      },
+      "size_bytes": 2437516
+    },
+    {
+      "file": "u2netp_vulkan_fp32.pte",
+      "precision": "fp32",
+      "methods": {
+        "forward": {
+          "inputs": [
+            {
+              "shape": [
+                1,
+                3,
+                320,
+                320
+              ],
+              "dtype": "float32"
+            }
+          ],
+          "outputs": [
+            {
+              "shape": [
+                1,
+                1,
+                320,
+                320
+              ],
+              "dtype": "float32"
+            }
+          ]
+        }
+      },
+      "size_bytes": 4646540
+    }
+  ]
+}

vulkan/u2netp_vulkan_fp16.pte

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c7c5697f204072f4d37269438a79578bb2893bf4cf0c1ebd871d9b20165793c
+size 2437516

vulkan/u2netp_vulkan_fp32.pte

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88523d0b54ec20e9129194f110955cbc6d2f5d90b2293c9a712636fb5d255212
+size 4646540