Jiangmiao Pang comments

Results 18 comments of


                                            Jiangmiao Pang

The reference image and key image are exactly the same

Hi @Hcayirli， we use images from detection set and tracking set for joint tracking. It is possible to use the same image as reference for training, especially for those images...

The reference image and key image are exactly the same

[Here](https://github.com/SysCV/qdtrack/blob/640e3668f3f29be820a878cdc7d14e9a535e53b2/qdtrack/datasets/coco_video_dataset.py#L21).

ValueError: need at least one array to concatenate

Hi all, please read the [README.md](https://github.com/open-mmlab/mmtracking/blob/master/configs/mot/deepsort/README.md) As these trackers are using offline trained detectors, you need to train the detector first and infer with the MOT model.

Do you support HOTA metric ?

Sorry that we haven't supported HOTA in this repo. You can refer to [this file](https://github.com/open-mmlab/mmtracking/blob/2c7a8af32f8d81069f40d87a368b23f26fbc3887/mmtrack/datasets/mot_challenge_dataset.py#L459) as a reference to modify the codebase by yourself. Or dump the outputs and eval...

SOT demo failed

According to your log, it seems that the Qt platform plugin on your machine has some problems. Please have a check. `This application failed to start because no Qt platform...

Hi, I've found a bug in sort_tracker.py

@leemengwei Thanks, we will have a check.

Can I train only tracker?

Hi, you can train the model in this way, which means load a pre-trained detector but freeze the parameters and only train the tracker. But in our experiments, joint training...

The model and loaded state dict do not match exactly

Thanks for the reporting. We will update the ckpt.