Is there a way to use document unwraping onnx with detection and recognition model?

[Dat0901]

I converted my UVDoc.pdiparams model to ONNX format using paddle2onnx, which gave me uvdoc.onnx.
However when I read throughI tools/infer/predict_system.py, I noticed there’s no existing code path to use uvdoc.onnx model. Currently I am using predict_system.py to run inference with onnx weight like this:

python3 /home/datdq/1WorkSpace/PaddleOCR_2x/tools/infer/predict_system.py --use_onnx=True \
 --det_model_dir="/home/user1/WorkSpace/PaddleOCR/output/onnx/PP-OCRv3_mobile_det.onnx" \
 --rec_model_dir="/home/user1/WorkSpace/PaddleOCR/output/onnx/en_PP-OCRv3_mobile_rec.onnx" \
 --image_dir="/home/Image/image1.jpg" \
 --det_limit_side_len=640 \
 --det_limit_type=min \
 --det_db_unclip_ratio=2.0 \
 --max_text_length=25 \
 --rec_image_shape="3,48,320" \
 --use_gpu=True \
 --rec_char_dict_path="/home/user1/WorkSpace/PaddleOCR/ppocr/utils/en_dict.txt" \
 --draw_img_save_dir="/home/user1/WorkSpace/PaddleOCR/inference_results" \

This works fine for detection and recognition, but I also want to enable document unwarping (UVDoc) in the same ONNX inference pipeline.
I know that using the Python API (paddleocr.PaddleOCR) allows document unwarping easily with::

from paddleocr import PaddleOCR  

ocr = PaddleOCR(
    use_doc_orientation_classify=False,
    use_doc_unwarping=True, # Enable document unwraping, which will download UVDoc.pdiparams
    use_textline_orientation=False,
    text_detection_model_name="PP-OCRv3_mobile_det",
    text_recognition_model_name="en_PP-OCRv3_mobile_rec",
    text_detection_model_dir="/home/user1/WorkSpace/PaddleOCR/output/PP-OCRv3_mobile_det/",
    text_recognition_model_dir="/home/user1/WorkSpace/PaddleOCR/output/en_PP-OCRv3_mobile_det/",
    text_det_limit_side_len=640,
    text_det_limit_type="min",
    text_det_unclip_ratio=2.0,
)

result = ocr.predict("/home/Image/image1.jpg")
for res in result:
    res.save_to_img("output")

But there’s no equivalent option in predict_system.py to load or use a UVDoc ONNX model.

So my questions are:

1. Is there any way to use document unwarping (uvdoc.onnx) inside tools/infer/predict_system.py, similar to how detection, recognition, and classification ONNX models are integrated?
2. Or am I missing an implementation step needed to run uvdoc.onnx together with other ONNX models?

[liuhongen1234567]

Hello, there are currently no plans to support UVDoc in predict_system.py. Since PaddleOCR 3.0, the involved modules have been significantly enhanced, and new models are integrated using from paddleocr import PaddleOCR. If you want to integrate UVDoc into predict_system.py, you can refer to the handling code in PaddleX for migration. The specific implementation code is as follows: https://github.com/PaddlePaddle/PaddleX/blob/release/3.2/paddlex/inference/models/image_unwarping/predictor.py
The usage documentation for document correction in PaddleX is as follows:
https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_image_unwarping.html#iii-quick-integration