Toward Generalizable Prompt Learning via Multi-Regularization Guided Knowledge Distillation.

Prompt learning has made significant progress in vision-language models (VLMs), enabling pre-trained models like CLIP to perform cross-domain tasks with few-shot or even zero-shot learning. However, existing methods tend to overfit the training data…

Preview only. Read the full abstract at the source