mikel-brostrom/boxmot

Unable to reproduce the benchmark results in README

wk565 opened this issue · 4 comments

wk565 commented

Search before asking

  • I have searched the Yolo Tracking issues and found no similar bug report.

Question

I used python3 tracking/val.py --imgsz 320 --classes 0 --benchmark MOT17-50 --yolo-model yolov8x.pt --reid-model osnet_x1_0_dukemtmcreid.pt --tracking-method bytetrack --verbose --source ./tracking/val_utils/data/MOT17-50/train，but the performance is very poor, the detector you use in the readme is the official yolov8X without any trainning?
The picture below is the performance results of bytetrack：
image

mikel-brostrom commented

For reproducing the old results you can set --conf 0.2. For reproducing the new results, download the detections and emebedding + second half of MOT17 from here:

https://github.com/mikel-brostrom/boxmot/releases/tag/v10.0.83

wk565 commented

For reproducing the old results you can set --conf 0.2. For reproducing the new results, download the detections and emebedding + second half of MOT17 from here:

https://github.com/mikel-brostrom/boxmot/releases/tag/v10.0.83

Is the detector used by the detections and emebedding you provided ultralytics' default yolov8X, or has it been trained on some datasets?

mikel-brostrom commented

You can find the detections and embeddings used for the new results here:

https://drive.google.com/drive/folders/1zzzUROXYXt8NjxO1WUcwSzqD-nn7rPNr

As you can see the detector used is YoloX-x and the embedding come from BoT. For more details read the paper 😄

  • 👍1
wk565 commented

You can find the detections and embeddings used for the new results here:

https://drive.google.com/drive/folders/1zzzUROXYXt8NjxO1WUcwSzqD-nn7rPNr

As you can see the detector used is YoloX-x and the embedding come from BoT. For more details read the paper 😄

Thanks，I thought it was the yolov8X's result，now I get it.

  • 👍1