search

MichiganCOG / ViP

Video Platform for Action Recognition and Object Detection in Pytorch
MIT License
219 stars 37 forks source link

Allow user to ignore final shape argument #22

Closed lemmersj closed 5 years ago

lemmersj commented 5 years ago

Currently train.py (and maybe eval.py?) checks that the final_shape argument matches the actual image returned from the dataloader (line 149). Some architectures are able to handle multiple input shapes. Providing a method of ignoring this assertion (perhaps by setting final_shape to -1) would be helpful in some cases.

lemmersj commented 5 years ago

I realized about thirty seconds after creating this issue why this was difficult, but it might be worth considering regardless.

zeonzir commented 5 years ago

You can switch minibatch size to 1 and pseudo batch loop to a high value. Since this condition isn't a natural case and variable mini-batch sizes isn't a common occurrence I would suggest you comment out the final shape assertion.

lemmersj commented 5 years ago

I think you're right. This issue is niche enough that the described workaround is acceptable.