Single ONNX for complete pipeline

[gautamHCSCV]

Can you please write a wrapper function which includes the entire flow from detection to recognition, into a single onnx model. This would also reduce the inference time and accuracy of the model.

Please provide me with code using which I can create a single onnx model which takes image as an input and provide text as output.

[SWHL]

As I know, multi onnx files can't be merged into one.
If there is relevant documentation, please let me know, I'd be more than happy to explore this approach.

[gautamHCSCV]

Yes multi onnx files can't be merged but we can merge multiple paddle files. I am not sure about this, but can we create a wrapper function which combines the entire pipeline into a single onnx model.

[SWHL]

I am a little confused about what it means to create a wrapper function which combines the entire pipleline into a single onnx model here.
Can you make a example?

[gautamHCSCV]

Meaning, is there a way we can have a single onnx model for complete paddle ocr model pipeline.

[SWHL]

The current models are all based on two-stage algorithms, which can't be merged a single model.
If you want to have only a single onnx model for OCR, you may need to see the End-to-End algorithm, such as PGNet. You can refer this link