Yolov5 support as a detector about yolo_tracking HOT 6 CLOSED

The problem here is that there is no Yolov5 pip package. I am trying to avoid submodules on this repo as much as possible. The only possibility I see is to add the Yolov5 repo as a submodule and create a detector using those resources 😄. I have however no plan of implementing this. Because again, submodules are a hassle to handle.

from yolo_tracking.

@mikel-brostrom That's very much understandable.

To implement what you have said, I have a hunch that there would be some custom modifications to be done to 'examples->detectors->yolo_interface.py/init.py' and a new file would need to be added for 'yolov5.py'? Is that direction right to think in?
I made some quick modifications but they didn't work as planned, so do you have some directions as to how I could get Yolov5 working? I only ask as it's very crucial for me to do this.
And again, thanks for such a quick response! :)

from yolo_tracking.

@mikel-brostrom - Another question that I wanted to ask here, is it possible for me to get the yolov5 output (from the Ultralytics repo) separately and just run the trackers (OC-SORT and DeepOCSORT) in boxmot on that output? What would be the way to do that?

And if the above technique works, I'd love to contribute it as a tutorial too (if it doesn't already exist) for others to -

1. get their output from their own detectors and
2. use boxmot's trackers on those outputs.

from yolo_tracking.

'examples->detectors->yolo_interface.py/init.py' and a new file would need to be added for 'yolov5.py'? Is that direction right to think in?

Yup that is right. But first you need the submodule for importing the right stuff such that you can load you yolov5 weights and run inference using it 😄

from yolo_tracking.

Yup that is right. But first you need the submodule for importing the right stuff such that you can load you yolov5 weights and run inference using it 😄

Got it.

@mikel-brostrom - Another question that I wanted to ask here, is it possible for me to get the yolov5 output (from the Ultralytics repo) separately and just run the trackers (OC-SORT and DeepOCSORT) in boxmot on that output? What would be the way to do that?

And if the above technique works, I'd love to contribute it as a tutorial too (if it doesn't already exist) for others to -

1. get their output from their own detectors and
2. use boxmot's trackers on those outputs.

Any thoughts on this?

from yolo_tracking.

Any thoughts on this?

You could store the detections for each frame in a list [dets_frame1, dets_frame2, ..., dets_frameN]. Then, you could dump this data such that you can reload it anywhere else. After that you could just run the examples provided by this repo 😄

from yolo_tracking.