search

yassouali / CCT

:page_facing_up: Semi-Supervised Semantic Segmentation with Cross-Consistency Training (CVPR 2020).
https://yassouali.github.io/cct_page/
MIT License
395 stars 58 forks source link

dataloader #65

Closed FrankWuuu closed 2 years ago

FrankWuuu commented 2 years ago

in dataloaderdrop_last=True? Otherwise, dataloader = iter(zip(cycle(self.supervised_loader), self.unsupervised_loader)) might get different size

yassouali commented 2 years ago

hi @FrankWuuu

drop_last=True just removes the last batch if the number of its example < batch size, so at most you might lose 9 examples (for a batch of 10) at a given iteration, but then in the next you will see these examples since we train with a shuffle, i don't think it'll have any significant impact; tho it shouldn't be used for validation