chrischoy/SpatioTemporalSegmentation

Possible bugs in train.py

thangvubk opened this issue · 1 comments

thangvubk commented

Thank you for a very nice project.
When scanning through lib/train.py, I found two possible bugs:

  1. L#78 coords[:, :3] += (torch.rand(3) * 100).type_as(coords) should be coords[:, 1:] += (torch.rand(3) * 100).type_as(coords) since first coords is the batch index
  2. L#83 input[:, 1:] = input[:, 1:] / 255. - 0.5 should be input[:, :3] = input[:, :3] / 255. - 0.5
chrischoy commented

Thanks for reporting the error. I've pushed the fix.