custom detection

[erkara]

Hi all

Thank you for this great library. I wonder if your library can be modified (can handle) to accept a custom Pytorch model trained to detect a single class?
I would super help my research thank you!

[mikel-brostrom]

I wonder if your library can be modified (can handle) to accept a custom Pytorch model trained to detect a single class?

Thir repo can load any Yolov5 model you have trained, for other architectures than Yolov5 you would need to customize the repo yourself

[erkara]

I am doing single class detection-tracking and that class is not one of COCO or others. Is there any tutorial for custom-datasets I can check out?

[mikel-brostrom]

https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data

[apaaris]

In that case, admitting that we train to detect a new category which is not in COCO, which StrongSORT weights should be used?
I would have a pretrained model which works for my custom category and I was looking on how to implement tracking, since ideally it's possible to load the yolo weights directly, I was wondering if there are any adjustments to be made to the StrongSORT part

[mikel-brostrom]

You could train a custom ReID model on your specific class for better ReID performance.