dvlab-research/ECCV22-P3AFormer-Tracking-Objects-as-Pixel-wise-Distributions

Training Sequence

smart-face opened this issue · 1 comments

smart-face commented

Verifying the results becomes challenging when insufficient information is present in the paper.
Could you please clarify the following training procedures?

  1. Why do you call coco pretrained? if your method consists of flow guided approach then do you need to train the entire model with center offset by using the coco dataset or did you just use the coco backbone model?
  2. If so, then is it the same for the crowdhuman and mot17 datasets?
    Please correct me if I am wrong.
smart-face commented

Could you please provide the supplementary materials of your work?