How to use mot17_fulltrain.pth model to track video?

[yandongwei]

When I use the below command : python demo.py tracking --load_model ../models/coco_tracking.pth --demo ../videos/test.avi It works well.

But when I want to use mot17_fulltrain.pth model ,I use this command : python demo.py tracking --load_model ../models/mot17_fulltrain.pth --demo ../videos/test.avi the output images all have no boxes ！

[biubug6]

The same problem!

[xingyizhou]

That's a good question. I haven't tested it yet but it is expected. In my point of view, the mot models are overfitted to street video surveillant scenes and not generalizable to general domains. You can try the crowdhuman model to see if it is better for your data.

[hampen2929]

It works well.

python demo.py tracking --load_model ../models/coco_tracking.pth --demo ../videos/video.mp4

I also tried the mot17_fulltrain and crowdhuman, and the output images have no boxes. Commands are below.

python demo.py tracking --load_model ../models/mot17_fulltrain.pth --demo ../videos/video.mp4
python demo.py tracking --load_model ../models/crowdhuman.pth --demo ../videos/video.mp4

[dutxubo]

I found the cause of this problem.
when I run:

python demo.py tracking --load_model ../models/crowdhuman.pth --demo ../videos/mytest.mp4

The output prompts that the weights and structure of the model do not match.

loaded ../models/crowdhuman.pth, epoch 140
Skip loading parameter hm.2.weight, required shapetorch.Size([80, 256, 1, 1]), loaded shapetorch.Size([1, 256, 1, 1]).
Skip loading parameter hm.2.bias, required shapetorch.Size([80]), loaded shapetorch.Size([1]).

Modify init function in opts.py

        #train_dataset = default_dataset_info[opt.task] \
        #  if opt.task in default_dataset_info else 'coco'
        train_dataset = opt.dataset

Then run

python demo.py tracking  --dataset crowdhuman --load_model ../models/crowdhuman.pth --demo ../videos/mytest.mp4 

It's OK.

[rose-jinyang]

Hello I want to know an exact command to test demo with the pre-trained model mot17_fulltrain.pth Thanks

[csdingithub]

I found the cause of this problem. when I run:

python demo.py tracking --load_model ../models/crowdhuman.pth --demo ../videos/mytest.mp4

The output prompts that the weights and structure of the model do not match.

loaded ../models/crowdhuman.pth, epoch 140
Skip loading parameter hm.2.weight, required shapetorch.Size([80, 256, 1, 1]), loaded shapetorch.Size([1, 256, 1, 1]).
Skip loading parameter hm.2.bias, required shapetorch.Size([80]), loaded shapetorch.Size([1]).

Modify init function in opts.py

        #train_dataset = default_dataset_info[opt.task] \
        #  if opt.task in default_dataset_info else 'coco'
        train_dataset = opt.dataset

Then run

python demo.py tracking  --dataset crowdhuman --load_model ../models/crowdhuman.pth --demo ../videos/mytest.mp4 

It's OK.

I modify the init function in opts.py. and then use the command. python demo.py tracking --dataset mot --load_model ../models/crowdhuman.pth --demo ../videos/mytest.mp4 I get some results,but it's not good!