Problem while deploying the newest official PP-OCRv5 on OpenVINO (no .pdmodel file found)

[lemonschwein]

Official PP-OCRv5 models doesn't have .pdmodel file anymore, how can I deploy the newest version of OCRv5 on OpenVINO.
All the blogs I found online relating to deploying PP-OCR models on OpenVINO is only for the PP-OCRv5 with .pdmodel files, but the new PP-OCR models replaces the .pdmodel file with .json file.

How can I solve such issue? I am trying the deploy the official PP-OCR models in PC with only using CPU and OpenVINO engine.
If anyone has a hint, please don't hestiate to share, thanks in advance! : )

[alex-dinh]

I just tried using theinference.json file in place of inference.pdmodel, and it was able to successfully execute this command to convert to onnx. Not sure if the onnx model actually works though. Perhaps this would also work for your use case?

paddle2onnx --model_dir PP-OCRv5_mobile_det_infer \
--model_filename inference.json \
--params_filename inference.pdiparams \
--save_file model.onnx \
--opset_version 11 \
--enable_onnx_checker True

[Bobholamovic]

PaddlePaddle 3.0 changes the model format from .pdmodel to .json, and OpenVINO does not currently support .json models. You can alternatively convert the .json model to the ONNX format (https://www.paddleocr.ai/latest/en/version3.x/deployment/obtaining_onnx_models.html) or use the PaddleOCR high-performance inference functionality (https://www.paddleocr.ai/latest/en/version3.x/deployment/high_performance_inference.html).