Abstract: Highlights•CoAPT is a novel prompt tuning method for few/zero-shot image classification.•CoAPT combines class attribute words and learnable prompts in a unified framework.•CoAPT can be easily incorporated into various existing prompt tuning methods.•A meta-network is used to generate input-image wise feature biases with soft prompts.•CoAPT leads to considerable improvements upon existing baseline methods.
External IDs:dblp:journals/kbs/LeeABL25
Loading