| ---
|
| library_name: onnxruntime
|
| pipeline_tag: object-detection
|
| license: mit
|
| base_model: microsoft/table-transformer-detection
|
| tags:
|
| - table-detection
|
| - table-structure-recognition
|
| - tatr
|
| - table-transformer
|
| - onnx
|
| - gmft
|
| ---
|
|
|
| # Microsoft Table Transformer (TATR) — ONNX re-export
|
|
|
| Re-export of Microsoft's Table Transformer to ONNX, packaged the way [GMFT](https://github.com/conjuncts/gmft) consumes it. Used by [ScanIndex](https://github.com/welcomyou/scanindex) for table detection + structure recognition during DOCX export.
|
|
|
| ## Variants
|
|
|
| | Subdir | Upstream | Task |
|
| |---|---|---|
|
| | `gmft_onnx/detection/model.onnx` | [`microsoft/table-transformer-detection`](https://huggingface.co/microsoft/table-transformer-detection) | Detect table bounding boxes on a page |
|
| | `gmft_onnx/structure/model.onnx` | [`microsoft/table-transformer-structure-recognition-v1.1-all`](https://huggingface.co/microsoft/table-transformer-structure-recognition-v1.1-all) | Detect rows / columns / cells inside a cropped table |
|
|
|
| Each subdir also contains the HF `config.json` + preprocessor metadata so `transformers` / `optimum` can wrap the ONNX directly.
|
|
|
| ## Loading
|
|
|
| ```python
|
| from huggingface_hub import snapshot_download
|
| local = snapshot_download("welcomyou/gmft-tatr-onnx", local_dir="models")
|
| # Detection: f"{local}/gmft_onnx/detection/model.onnx"
|
| # Structure: f"{local}/gmft_onnx/structure/model.onnx"
|
| ```
|
|
|
| ## Re-export reproduction
|
|
|
| See [train-convert/gmft/convert/export_gmft_tatr_to_onnx.py](https://github.com/welcomyou/scanindex/blob/main/train-convert/gmft/convert/export_gmft_tatr_to_onnx.py).
|
|
|
| ## License
|
|
|
| MIT, inherited from Microsoft Table Transformer.
|
|
|
