classifier-balancing
classifier-balancing copied to clipboard
Question about weight norm
Dear experts or scholars:
Thanks for your excellent work on long tail recognition.Your paper is now inevitable to read and I have read many times.
I have a question about weight norm. In figure 2 left, you have show us weights are closely related to class index or numbers of images of classes. My question is how can you find the relationship between classes and weight norm?Or how can you get the weight norm of one class?
I'm thinking it's probably something like this -
torch.norm(W, 2, 1) // This gives K-length tensor corresponding to norm of each class