Feature Request: YOLO Classifier Integration

[nghiaanh2019]

Hi team,
I would like to request adding support for YOLO-based image classification (YOLO Classifier) in our project. Currently, we have YOLO for object detection, but there is a need to classify images (e.g., sit/stand, object state) using YOLO's classifier head.
Integrate YOLO classification models (e.g., yolov8-cls) as a secondary GIE in DeepStream.

[IJLee0812]

Hi team, I would like to request adding support for YOLO-based image classification (YOLO Classifier) in our project. Currently, we have YOLO for object detection, but there is a need to classify images (e.g., sit/stand, object state) using YOLO's classifier head. Integrate YOLO classification models (e.g., yolov8-cls) as a secondary GIE in DeepStream.

Hello nghiaanh,

I'm in a similar situation.

I've successfully managed to port the standard yolov8m model. However, I understand that to build a classification model, I need an ONNX export script and a C++-based custom parser to interpret the output of the classification model.

For reference, someone in a pull request on the repository attempted this with a YOLOv5 version. I tried to apply it, but it doesn't work correctly, likely due to version compatibility issues. You might want to take a look.

Also, I was wondering if you have any experience with fine-tuning your YOLOv8 detection model.
If you have, did you use the yolov8_export.py and the custom parser.so files as-is and only modify the num-detected-classes parameter in nvinfer?

I would appreciate your response. I really appreciate the effort you're putting in.