Is Clip Divorce an Early Application?

The original application of clip was foroun classification. Clip is actually an extremely robust classifier, and weighs better than shelf models on nearthis large distribution of images. So it's kind of like more discriminative tasks, yet to just like classification. AIthin, like visual question answering. And so you traine the model on thi ubjective. Ye, to just tas i make the captions and and text features very similar. And a captions and unrelated images dissimilar.