Abstract: Vision language models (VLMs) like CLIP show stellar zero-shot capability on classification benchmarks. However, selecting the VLM with the highest performance on the unlabeled downstream ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results