Closed grypesc closed 9 months ago
Hi, our method SimGCD is introduced in the section 4 of our paper, and from the section 4.1 and section 4.2, you can see the four losses we used to train the model.
Thanks for clarification, I misunderstood that one
Hi, in your paper you state that you use two loss functions to train the feature extractor (SupCL and SelfSupCL). But in your code I can see 4 losses where there is a cross entropy loss between targets and outputs of the feature extractor:
This loss also seems to have effect on the feature extractor, can you elaborate?