Research map / Architectures and efficiency / Neural architecture search
CM-NAS: Cross-Modality Neural Architecture Search for Visible-Infrared Person Re-Identification
Chaoyou Fu, Yibo Hu, Xiang Wu, Hailin Shi, Tao Mei, Ran He
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| search BN separation scheme instead of fixed two-stream (BN) | CM-NAS (basic loss) · SYSU-MM01 all-search single-shot | Rank-1/mAP: 2.62% and 1.55% | Improved |
| add C3MMD loss (ablation) | two-stream (BN) · SYSU-MM01 all-search single-shot | Rank-1/mAP: 1.0% and 1.06% | Improved |
| add C3MMD loss (ablation) | one-stream · SYSU-MM01 all-search single-shot | Rank-1: 54.49 -> 56.18 | Improved |
| replace BN with Instance Normalization (ablation) | one-stream · SYSU-MM01 all-search single-shot | Rank-1: 54.49 -> 45.54 | Hurt |
| separate blocks of stage1 and stage2 into two streams (ablation) | one-stream · SYSU-MM01 all-search single-shot | Rank-1: 54.49 -> 56.20 | Improved |
| separate only BN layers rather than entire blocks (ablation) | two-stream · SYSU-MM01 all-search single-shot | Rank-1: 56.20 -> 56.94 | Improved |
| use C3MMD in both search and training phases rather than phase 1 only (ablation) | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1: 59.55 -> 61.99 | Improved |
| add CMMD term to C3MMD (ablation) | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1: 61.99 -> 60.77 | Improved |
| add correlation consistency term CC to C3MMD (ablation) | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1: 61.99 -> 60.27 | Improved |
| use class-specific MMD instead of plain MMD (ablation) | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1: 61.99 -> 60.83 | Improved |
| use BN-oriented search space instead of Auto-ReID search space | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1/mAP: 16.48%/13.96% | Improved |
| search architecture on RegDB then train on SYSU-MM01 | CM-NAS · SYSU-MM01 all-search single-shot | Rank-1: 61.99 -> 60.42 | Hurt |
Closest papers by meaning
- Cross-Modality Transformer for Visible-Infrared Person Re-identification (ECCV 2022 · similarity 0.94)
- Cross-Modality Person Re-identification with Memory-Based Contrastive Embedding (AAAI 2023 · similarity 0.92)
- Syncretic Modality Collaborative Learning for Visible Infrared Person Re-Identification (ICCV 2021 · similarity 0.92)
- Shallow-Deep Collaborative Learning for Unsupervised Visible-Infrared Person Re-Identification (CVPR 2024 · similarity 0.91)
- Implicit Discriminative Knowledge Learning for Visible-Infrared Person Re-Identification (CVPR 2024 · similarity 0.91)
- Empowering Visible-Infrared Person Re-Identification with Large Foundation Models (NeurIPS 2024 · similarity 0.90)
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.
