How do you handle with the loss_match of track_head in QueryTrack?

[kay-Lifeng]

Dear author：
In MaskTrack, when calculate the loss_match of track_head, the picture having only one instance will not participate in the calculation of loss_mach. So, if the pictures of one batch all have only one instance, the loss_match of this batch will be 0. How do you solve this problem in your QueryTrack?

[Yuxin-CV]

Please refer to the implantation of Mask Track R-CNN.

[Yuxin-CV]

I'm closing this issue but let us know if you have any further questions.