How to use own detection model like YOLOV8

[nikhilgarg1210]

Bug description

Hi @felixdittrich92, i want to use yolov8 detection model instead of the prebuilt models by you. if yes how can i use them. could you please help me.

Code snippet to reproduce the bug

# Load the ONNXTR OCR model
def load_onnxtr_model():
    return ocr_predictor(
        det_arch='db_resnet50', #i need to use yolov8
        reco_arch='crnn_vgg16_bn',
        detect_orientation=True,
        straighten_pages=True,
        assume_straight_pages=False,
    )

Error traceback

nothing

Environment

python3

[felixdittrich92]

Hi @nikhilgarg1210 👋;

Currently it's not supported to load a yolo exported model for text detection :)

Do you need only the recognition part from OnnxTR or also the rest like image straighten and the DocumentBuilder features ?

We have an contribution module where you can load an Yolo Onnx exported model (no OBB yet) https://mindee.github.io/doctr/using_doctr/using_contrib_modules.html
but that was more meant for artefact detection models ^^