search

ifzhang / FairMOT

[IJCV-2021] FairMOT: On the Fairness of Detection and Re-Identification in Multi-Object Tracking
MIT License
4k stars 934 forks source link

0 detection when running demo.py with *dla* pre-trained models on custom videos. #247

Open mainguyenanhvu opened 4 years ago

mainguyenanhvu commented 4 years ago

I found that there was 0 detection when I run demo.py with dla pre-trained models on my custom videos.

I wonder that whether or not I need to train them again and re-run demo.py. I hope to receive your help.

Thank you.

ifzhang commented 4 years ago

There may exist a domain gap between the custom videos and our training videos. You can first adjust --conf_thresh to a low score like 0.1 and see whether the model can detect some people. If it still has 0 detection, you need to finetune the model on your custom videos.

bolin-chen commented 3 years ago

I met the same situation. Even if if adjust conf_thresh to 0.1, it still cannot detection any person. I have tried several videos, some of which contains obvious people. But I just get the same results. It seems that it is not a data domain problem.