Dayan Wu

Beyond Post-Quantization: Native Hash Learning with a Dedicated HASH Token

Xinze Liu, Ding Wang, Dayan Wu, Hengjie Zhu, Peng Fu, Zheng Lin, et al.

Efficient large-scale image retrieval requires compact representations that preserve semantic similarity under fast Hamming-space search. Deep hashing is appealing, but most existing CNN- and ViT-based methods still follow a post-quantization paradigm, where continuous visual fea…Read more