1 paper
Thang M. Pham, Peijie Chen, Tin Nguyen +3
CLIP-based classifiers rely on the prompt containing a {class name} that is known to the text encoder. Therefore, they perform poorly on new classes or the classes whose names rare…