1 paper
Aotian Zheng, Winston Sun, Bahaa Alattar +2
CLIP-based person re-identification (ReID) methods aggregate spatial features into a single global \texttt{[CLS]} token optimized for image-text alignment rather than spatial selec…