Fetching the paper…

Does CLIP's Generalization Performance Mainly Stem from High Train-Test Similarity? · Around