Evaluation on MOT17 dataset

[navaravan]

Hi @xingyizhou

Thanks for sharing the code How did you evaluate the tracker on MOT17 dataset and got the results. Can you list the steps followed for recreating the results. Thanks in advance.

[xingyizhou]

Please check here.

[Huwenzjw]

Thanks for your answer. These are some steps about how to train on MOT17. Where can I find some steps about testing on MOT17.

[akamboj2]

@Huwenzjw

I think the instructions linked describe how to evaluate as well. Near the top of that link it says: "To evaluate a model with a trained/ pretrained model, run python train_net.py --config-file /path/to/config/name.yaml --eval-only MODEL.WEIGHTS /path/to/weight.pth "

I downloaded the model, placed it in models/ than ran: python train_net.py --config-file configs/GTR_MOT_FPN.yaml --eval-only MODEL.WEIGHTS models/GTR_MOT_FPN.pth

and I'm getting the same metrics they reported. MOTA 71.253 IDF1 75.902 HOTA 63.047 etc...

The only caveat is the first time i ran it, I was getting an error saying it can't find datasets/mot/MOT17/seqmaps/trainval.txt, but i saw that file in gtr/tracking/seqmaps/ so i just created a symlink (I went to datasets/mot/MOT17/ directory and did ln -s GTR_ROOT/gtr/tracking/seqmaps seqmaps) and then it worked fine.

[gurkirt]

I was able to train on MOT17 including a detector. I got similar results but not exactly the same result because I could only fit a sequence length of 6 rather than 8 at training time.

[wlc17662354484]

@xingyizhou 想问一下怎么样检测在测试集上的效果，我按照GTR_MOT_FPNfull.yaml以及相应的权重文件进行测试的时候，就会显示 这个错误