SupConLoss on non-image data (no augmentation)

HobbitLong / SupContrast

PyTorch implementation of "Supervised Contrastive Learning" (and SimCLR incidentally)

BSD 2-Clause "Simplified" License

3.06k stars 526 forks source link

SupConLoss on non-image data (no augmentation) #87

Open mese79 opened 3 years ago

mese79 commented 3 years ago

Hi
Thanks for your great paper and code!
I wonder is this loss effective on non-image data?
I mean I can't produce augmented version of my data, but positive and negative pairs on each batch.

htn274 commented 2 years ago

Hi, I have the same interest in applying this model to non-image data. Did you make any experiments?