CONWISE-FT: ENHANCING FEW-SHOT LEARNING THROUGH CONTRASTIVE PROJECTION TUNING AND POST-TRAINING WEIGHT ENSEMBLING

Foundation vision–language models such as CLIP achieve strong zero-shot performance, however adapting them to specialised few-shot tasks still remains difficult. Two failure modes recur: models tend to overfit to the few samples given during training if the process is not carefully designed, and als