发布 1.0.1 PicoDet FP32 WebGPU 候选

1.0.1/README.md

@@ -0,0 +1,17 @@
+# PP-PicoDet-L-320 1.0.1
+
+该版本目前处于 `labs/blocked`。目录包含经过 WebGPU 兼容清理的 FP32 ONNX
+候选 `picodet-l-320-fp32.onnx`，但三类不可变公共来源、WASM 和真实 NVIDIA
+WebGPU 浏览器证据尚未全部完成，因此不能作为稳定默认模型使用。
+
+候选模型大小为 `23243834` 字节，SHA-256 为
+`0397bb449689d1bf57dfcb8849b3ddaa1c8962e1e63e533bd97d265908a428a1`。
+清理只将 PicoDet 四个共享 `MatMul_0..3` 的 `auto_4_` 权重从 `[8]` 规范为
+`[8,1]`，规避 ORT WebGPU 的 2D×1D MatMul 限制；输入输出契约、opset 和参数
+数量保持不变。CPU parity 和清理结构证据分别见：
+
+- `tools/model-pipeline/reports/1.0.1/fp32-cpu-parity.json`
+- `tools/model-pipeline/reports/1.0.1/sanitize.json`
+
+FP16 资产尚未在本版本生成。来源 revision、下载地址、字节数和 SHA-256 在
+完成公共上传与回读核验后再写入 manifest；不要把本地文件直接当作已发布来源。

1.0.1/labels/coco.txt

@@ -0,0 +1,80 @@
+person
+bicycle
+car
+motorcycle
+airplane
+bus
+train
+truck
+boat
+traffic light
+fire hydrant
+stop sign
+parking meter
+bench
+bird
+cat
+dog
+horse
+sheep
+cow
+elephant
+bear
+zebra
+giraffe
+backpack
+umbrella
+handbag
+tie
+suitcase
+frisbee
+skis
+snowboard
+sports ball
+kite
+baseball bat
+baseball glove
+skateboard
+surfboard
+tennis racket
+bottle
+wine glass
+cup
+fork
+knife
+spoon
+bowl
+banana
+apple
+sandwich
+orange
+broccoli
+carrot
+hot dog
+pizza
+donut
+cake
+chair
+couch
+potted plant
+bed
+dining table
+toilet
+tv
+laptop
+mouse
+remote
+keyboard
+cell phone
+microwave
+oven
+toaster
+sink
+refrigerator
+book
+clock
+vase
+scissors
+teddy bear
+hair drier
+toothbrush

1.0.1/manifest.json

@@ -0,0 +1,15 @@
+{
+  "schemaVersion": 1,
+  "status": "labs/blocked",
+  "model": {
+    "id": "pp-picodet-l-320",
+    "version": "1.0.1",
+    "architecture": "PicoDet-L-320 LCNet",
+    "format": "onnx"
+  },
+  "variants": [],
+  "blocked": {
+    "reason": "1.0.1 清理、FP32/WASM、FP16 和真实 WebGPU 证据尚未完成。",
+    "evidence": "tools/model-pipeline/reports/1.0.1/"
+  }
+}

1.0.1/picodet-l-320-fp32.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0397bb449689d1bf57dfcb8849b3ddaa1c8962e1e63e533bd97d265908a428a1
+size 23243834

1.0.1/pipeline.json

@@ -0,0 +1,19 @@
+{
+  "schemaVersion": 1,
+  "status": "labs/blocked",
+  "modelVersion": "1.0.1",
+  "source": "models/pp-detection/1.0.0/picodet-l-320-fp32.onnx",
+  "target": "picodet-l-320-fp32.onnx",
+  "steps": [
+    "sanitize_picodet_webgpu_fp32.py",
+    "onnx.checker.check_model",
+    "onnx.shape_inference.infer_shapes",
+    "CPU parity on seven fixtures"
+  ],
+  "webgpuCompatibility": {
+    "matmulWeight": "auto_4_",
+    "beforeShape": [8],
+    "afterShape": [8, 1]
+  },
+  "blockedReason": "公共来源、WASM、真实 NVIDIA WebGPU 和 FP16 证据尚未全部完成"
+}

1.0.1/source.yaml

@@ -0,0 +1,17 @@
+status: labs/blocked
+version: 1.0.1
+reason: FP32 WebGPU 清理候选已生成，但三类不可变来源、WASM、真实 NVIDIA WebGPU 和 FP16 证据尚未完成，禁止生成 stable 变体。
+artifacts:
+  fp32:
+    path: picodet-l-320-fp32.onnx
+    bytes: 23243834
+    sha256: 0397bb449689d1bf57dfcb8849b3ddaa1c8962e1e63e533bd97d265908a428a1
+    precision: fp32
+    quantization: none
+    opset: 11
+    parameterCount: 5787988
+  fp16:
+    status: pending
+sources:
+  status: pending
+  required: [huggingface, modelscope, git-lfs]