Performance evaluation on MOT20 seems abnormal

GeekAlexis / FastMOT

High-performance multiple object tracking based on YOLO, Deep SORT, and KLT 🚀

MIT License

1.15k stars 253 forks source link

Performance evaluation on MOT20 seems abnormal #244

Closed PonyMaY closed 2 years ago

PonyMaY commented 2 years ago

Hi Alex! Recently I check the FastMOT peformance on MOT20 by inputing MOT20 datasets from MOTchallenge. But this seems that i can't get your result by using the data in /eval/results/MOT20-0x.txt you provided. MOT20 My result is shown above. The result of MOT20-03 and MOT20-05 are extremely abnoraml. Thanks in advance!

GeekAlexis commented 2 years ago

Many people have already replicated my results. See #85. You probably did not set the frame size correctly in mot.json for MOT20-03 and MOT20-05.