clipmulti-label

  • Adapter is 3 layer FFN with Relu and dropout

Use CLIP model to train model on text data and transfer to image domain

Ideas

  • LiT tuning of language model first for new domain