| --- |
| license: apache-2.0 |
| --- |
| |
| # Introduction |
| This repository hosts [PaddleOCR PP-DocLayoutV3](https://huggingface.co/PaddlePaddle/PP-DocLayoutV3_safetensors), |
| an RT-DETR-based **document layout detector** (~33M params), for the |
| [React Native ExecuTorch](https://www.npmjs.com/package/react-native-executorch) library, |
| exported to `.pte` for the **ExecuTorch** runtime (XNNPACK, CoreML, Vulkan). It finds and |
| classifies document regions — titles, paragraphs, tables, figures, formulas, headers/footers, |
| etc. — and is a companion to |
| [`react-native-executorch-paddleocr`](https://huggingface.co/software-mansion/react-native-executorch-PP-OCRv6). |
|
|
| If you'd like to run these models in your own ExecuTorch runtime, refer to the |
| [official documentation](https://pytorch.org/executorch/stable/index.html) for setup instructions. |
|
|
| The `.pte` is a pure tensor→tensor function; all pre/post-processing (resize, normalize, score |
| threshold, box convert) is the client's job. |
|
|
| ## Output contract |
|
|
| A single static method `forward`, fixed input (no buckets): |
|
|
| ``` |
| in [1, 3, 800, 800] # RGB, ImageNet-normalized (x/255 - mean)/std |
| out logits [1, 300, 25] # 25 layout classes, per query (apply sigmoid) |
| pred_boxes [1, 300, 4] # (cx, cy, w, h), normalized [0,1] |
| ``` |
|
|
| PP-DocLayoutV3 is a **DETR set-prediction** model → **no NMS**. Post-processing is just: |
| `score = sigmoid(logits)`, keep queries above a threshold, convert `(cx,cy,w,h) → (x1,y1,x2,y2)`, |
| scale to image size. Class names are in `labels.json` (index → label). |
|
|
| ### Classes (25) |
|
|
| `abstract, algorithm, aside_text, chart, content, formula, doc_title, figure_title, footer, |
| footnote, formula_number, header, image, number, paragraph_title, reference, reference_content, |
| seal, table, text, vision_footnote` (some indices map to the same display label; use |
| `labels.json` as the authoritative index→label map). |
| |
| ## Backends, sizes & latency (warm) |
| |
| | backend | target | precision | size | latency | |
| |---|---|---|---|---| |
| | `xnnpack` | CPU | fp32 | 132 MB | ~2.0 s (S24) | |
| | `coreml` | Apple ANE | fp16 | 91 MB | ANE fp16 | |
| | `vulkan` | Android GPU | fp16 (mixed-delegate) | **66 MB** | **~0.86 s (S24)** | |
| |
| > **Vulkan is the recommended Android backend** — ~2.4× faster than XNNPACK and half the size. |
| > It's mixed-delegate: most of RT-DETR runs fp16 on the GPU, while the box-head matmuls run on |
| > XNNPACK (they delegate as `addmm`→`linear`). XNNPACK stays fp32 because RT-DETR's deformable |
| > attention feeds non-contiguous tensors that int8/portable paths mis-handle. |
| |
| |
| |
| ## Compatibility |
| |
| If you intend to use these models outside of React Native ExecuTorch, make sure your runtime is |
| compatible with the **ExecuTorch** version used to export the `.pte` files. For more details, see |
| the compatibility note in the |
| [ExecuTorch GitHub repository](https://github.com/pytorch/executorch/blob/main/runtime/COMPATIBILITY.md). |
| If you work with React Native ExecuTorch, the library constants guarantee compatibility with the |
| runtime used behind the scenes. |
| |
| |
