Qingpeng Li

If there are no more comments about the implementation, I will add some tests.

I am currently working on fixing softmax focal algorithm in https://github.com/open-mmlab/mmcv/pull/2893 as an outside contributor. Per my understanding, the current sigmoid focal loss implementation is correct. It is very similar...

> as you can see the results between torchvision implementation and this are different the num_classes=1 because size of the logits last dim=1 for binary/focal classification torchvision implementation is correct...