train.py

[BudBudding]

Your work is very interesting and has research value and significance. Can you provide the complete code on your end? Including train.py and others.

[mannatsingh]

Hi @Eliauk-TiAmo thanks, and sorry for the late response, the repository was mistakenly archived!

The implementation for SWAG should be fairly straightforward, you just need to convert the multi-label targets to a probability by normalizing by the number of positive labels, and then apply cross entropy loss.

https://github.com/facebookresearch/omnivore has code for training a ViT with cross entropy loss. Unfortunately we won't be able to provide more detailed support with this.