PaddlePaddle

/

PP-DocLayoutV3

like 75

Follow

PaddlePaddle 2.57k

Great work！

Hi @alex-dinh , how did you convert the model to onnx. Using paddlex --paddle2onnx fails. Did you have to modify any operator?

Thanks!

I detailed how to do the conversion in this PR here: https://github.com/PaddlePaddle/Paddle2ONNX/pull/1619

Hi, does your version enable batching? Currently, I can't batch on the native version because the authors hardcoded batch=1 as a literal integer in the torch source's attention/decoder
reshapes, and torch.onnx.export bakes those literals into frozen graph constants.

Hi @xDido , yes batching is supported. Here is an example python script for batched inference: https://huggingface.co/alex-dinh/PP-DocLayoutV3-ONNX/blob/main/batched_inference_example.py