search

Shengyu-Feng / DSG-DETR

6 stars 1 forks source link

Output shape different for train and eval #2

Open Saurabh1409 opened 1 year ago

Saurabh1409 commented 1 year ago

Why is the shape of global output different during training and evaluation for the same input video? I have passed the same vid input to train.py and test.py, but the global output shape differs for both of these cases. _vid_id : 01KML global_output (train): torch.Size([38, 1936]) globaloutput (test): torch.Size([272, 1936]) I am unable to understand this requirement. Can you please help me with this?

Shengyu-Feng commented 1 month ago

Sorry for the late reply (the GitHub did not forward the issue to my email). We do some downsampling during the training time while the testing is conducted over the full video sequences.