English

Deep Robust Multilevel Semantic Cross-Modal Hashing

Computer Vision and Pattern Recognition 2020-10-07 v2

Abstract

Hashing based cross-modal retrieval has recently made significant progress. But straightforward embedding data from different modalities into a joint Hamming space will inevitably produce false codes due to the intrinsic modality discrepancy and noises. We present a novel Robust Multilevel Semantic Hashing (RMSH) for more accurate cross-modal retrieval. It seeks to preserve fine-grained similarity among data with rich semantics, while explicitly require distances between dissimilar points to be larger than a specific value for strong robustness. For this, we give an effective bound of this value based on the information coding-theoretic analysis, and the above goals are embodied into a margin-adaptive triplet loss. Furthermore, we introduce pseudo-codes via fusing multiple hash codes to explore seldom-seen semantics, alleviating the sparsity problem of similarity information. Experiments on three benchmarks show the validity of the derived bounds, and our method achieves state-of-the-art performance.

Keywords

Cite

@article{arxiv.2002.02698,
  title  = {Deep Robust Multilevel Semantic Cross-Modal Hashing},
  author = {Ge Song and Jun Zhao and Xiaoyang Tan},
  journal= {arXiv preprint arXiv:2002.02698},
  year   = {2020}
}

Comments

11 pages, 9 figures, submitted to a journal

R2 v1 2026-06-23T13:34:03.106Z