|
--- |
|
license: mit |
|
tags: |
|
- table-transformer |
|
- table detection |
|
- table structure recognition |
|
- int8 |
|
- Intel® Neural Compressor |
|
- neural-compressor |
|
- PostTrainingStatic |
|
- onnx |
|
datasets: |
|
- bsmock/pubtables-1m |
|
--- |
|
# INT8 Table Transformer |
|
|
|
## Post-training static quantization |
|
|
|
### ONNX |
|
|
|
This repo contains the models for: |
|
|
|
1) Table detection, |
|
2) Table structure recognition, |
|
|
|
The original FP32 PyTorch model comes from [bsmock/tatr-pubtables1m-v1.0](https://huggingface.co/bsmock/tatr-pubtables1m-v1.0). The INT8 ONNX models are quantized with [Intel® Neural Compressor](https://github.com/intel/neural-compressor). |
|
|
|
Refer to this [link](https://github.com/intel/neural-compressor/tree/master/examples/onnxrt/object_detection/table_transformer/quantization/ptq_static) for model preparation, quantization and benchmark scripts. |
|
|
|
#### Test result |
|
|
|
Table detection: |
|
|
|
| |INT8|FP32| |
|
|---|:---:|:---:| |
|
| **COCO metrics (AP)** |0.9691|0.9706| |
|
| **Model size (MB)** |56|111| |
|
|
|
Table structure recognition： |
|
| |INT8|FP32| |
|
|---|:---:|:---:| |
|
| **Model size (MB)** |56|111| |
|
|
