Research map / Label-efficient and robust learning / Recognition and long-tail
Farewell to Mutual Information: Variational Distillation for Cross-Modal Person Re-Identification
Xudong Tian, Zhizhong Zhang, Shaohui Lin, Yanyun Qu, Yuan Xie, Lizhuang Ma
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the full VSD, VCD, and VML method instead of the three-branch baseline (ablation) | Cross-modal person Re-ID · SYSU-MM01 all-search single-shot | Rank-1: 48.82 -> 60.02 | Improved |
| Add information bottleneck branches without variational distillation (ablation) | Three-branch Re-ID baseline · SYSU-MM01 all-search single-shot | Rank-1: 48.82 -> 54.85 | Improved |
| Replace variational distillation with conventional IB (ablation) | Three-branch Re-ID model · SYSU-MM01 all-search single-shot | Rank-1: 60.02 -> 8.79 | Hurt |
| Add VSD to the modality-specific branches (ablation) | Three-branch Re-ID model · SYSU-MM01 all-search single-shot | Rank-1: 54.85 -> 59.62 | Improved |
| Add VCD and VML to the modality-shared branch (ablation) | VSD Re-ID model · SYSU-MM01 all-search single-shot | Rank-1: 59.62 -> 60.02 | Improved |
| Use the full variational distillation method rather than the strongest listed baseline | Cross-modal person Re-ID · RegDB visible-to-thermal | Rank-1: 70.9 -> 73.2 | Improved |
| Use the full variational distillation method rather than the strongest listed baseline | Cross-modal person Re-ID · RegDB thermal-to-visible | Rank-1: 68.1 -> 71.8 | Improved |
| Avoid the mutual-information estimator used by conventional IB | Re-ID training · Training time | training time: 1.26x -> 1.09x | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Hi-CMD — Hi-CMD: Hierarchical Cross-Modality Disentanglement for Visible-Infrared Person Re-Identification (CVPR 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Variational Distillation for Multi-View Learning (TPAMI 2023 · similarity 0.88)
- Interact, Embed, and EnlargE: Boosting Modality-Specific Representations for Multi-Modal Person Re-identification (AAAI 2022 · similarity 0.87)
- Disentangled Cross-Modal Representation Learning with Enhanced Mutual Supervision (NeurIPS 2025 · similarity 0.86)
- Learning Memory-Augmented Unidirectional Metrics for Cross-Modality Person Re-Identification (CVPR 2022 · similarity 0.86)
- Human-centered Interactive Learning via MLLMs for Text-to-Image Person Re-identification (CVPR 2025 · similarity 0.85)
- RedCore: Relative Advantage Aware Cross-Modal Representation Learning for Missing Modalities with Imbalanced Missing Rates (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
