Adding custom text_rec onnx model to the StructureV3 pipeline

[fvapolina]

I have custom .onnx model for text recognition.

I'd like to just put the path to it into my PP-StructureV3.yaml config and get markdown at the end. Like this:

Unfortunately model_dir expects to have a directory with paddle format model, not an .onnx model.

I also attempted to convert .onnx model into paddle formatted model with X2Paddle, but got errors during the conversion so it did not work.

Question:
What are my options to implement custom .onnx text rec model into the existing StructureV3 pipeline?

[Bobholamovic]

You can try renaming the model file to inference.onnx.