segment-anything-with-clip
segment-anything-with-clip copied to clipboard
counter-examples
Hi! Do you think it's possible to modify the code to add images counter-examples in the support set?
Could you explain more about counter example?
By the way, PullRequest is warmly welcomed.
Sorry, I should have provided more context, but I was wondering if, in order to increase performance, it would be useful to provide images of other categories. For example, we're looking for dogs of a specific breed and the CLIP model isn't necessarily powerful enough to put a name to that breed. We could provide images of other breeds of dog in order to differentiate them.
I think it's a possible approach. We can use the clip image embedding instead of text embeddings.