guillaumeerhard / supervised_contrastive_loss_pytorch Goto Github PK

View Code? Open in Web Editor NEW

71.0 71.0 11.0 942 KB

Independent implementation of Supervised Contrastive Loss. Straight to the point and beyond

License: MIT License

Python 100.00%

cifar10 contrastive-learning deep-learning image-classification

supervised_contrastive_loss_pytorch's People

Contributors

Stargazers

Watchers

Forkers

hanzhang-wang 32nguyen zhangjiwei-japan guoyang-xie magnumenforcer bruno686 siriushou mojeedoyedeji imaodong shayandaneshvar luzhoushili

supervised_contrastive_loss_pytorch's Issues

Two different Data augmentations?

Hi,
According to the theory, in stage 1, two different types of augmentations should be applied to a batch of samples.

crop and resize
AutoAugment/randaugment/simaugment.
It seems the code doesn't have augmentation number (2). Are the mentioned results from only augmentation number (1) which has been applied twice to the same set of samples?

Thanks

Low Val Accuracy on CIFAR 100

Do you mind sharing the parameters you used for training CIFAR 100? For me validation accuracy seems to be saturating at 40% (using resnet-20 though) even with Auto Augment=True. I am interested in knowing the learning rate, batch size, resnet model and temperature you used for CIFAR-100.

Recommend Projects