Can someone please tell me if I have a different number of images per class during training on how to tackle the class imbalance problem in PyTorch?
How to solve this “class imbalance problem” in PyTorch?
Can someone please tell me if I have a different number of images per class during training on how to tackle the class imbalance problem in PyTorch?
How to solve this “class imbalance problem” in PyTorch?
You could over/undersample the classes using a WeightedRandomSampler
.
Have a look at this post for a small example.