Lei Ke

RTX gpu and less than 5 hours (in my impression)

You can compute it offline on the validation set. Doing it by modifying this [python file](https://github.com/SysCV/pcan/blob/main/pcan/datasets/coco_video_dataset.py#L307) adding the following code section similar to following at L307: > super_metrics = ['bbox',...

In MOTS, we consider more MOTA metric becuase single-frame mask ap doesnot consider the object tracking accuracy. The mask ap is YTVIS is also a tube-mask ap (including the object...

Hi, you can modify the dataloader [here](https://github.com/SysCV/pcan/blob/06416f1c96b7a86754828582d9a95b9ce0d327ba/pcan/apis/test.py#L23) to let it read image sequence only from the video directories w/o reading the corresponding GT labeling. Or you can generate fake annotation...

Hi, I have not tried this. For current code version, as we made a deep modification to original code in some python files, you need to make sure the modification...

what do you mean by 'video memory'? what's the backbone you are using?

You can refer to the discussion [here](https://github.com/SysCV/transfiner/issues/36).

you can set the class number to 1 and delete the original 80 classes embedding weights. Or you set the class index of the human dataset the same as coco...

The quickest way is to make you human dataset in coco format. You can refer to more details [here](https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html).

"after fix random_color() , other errors still occurred." What are the other errors that you met?