Research map / Architectures and efficiency / Attention and transformers
TransMatcher: Deep Image Matching Through Transformers for Generalizable Person Re-identification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace ViT / vanilla Transformer with TransMatcher simplified decoder | Transformer image matcher · MSMT17 -> CUHK03-NP/Market-1501 | mAcc (%): 27.42 -> 44.29 | Helped |
| concatenate query and gallery features in ViT (Transformer-Cat) | ViT · MSMT17 cross-dataset | mAcc (%): 27.42 -> 25.34 | No clear effect |
| feed query feature maps into decoder instead of learnable query embeddings (Transformer-Cross) | vanilla Transformer · MSMT17 cross-dataset | mAcc (%): 27.01 -> 36.70 | Helped |
| use TransMatcher simplified decoder instead of Transformer-Cross at same size | Transformer-Cross · MSMT17 cross-dataset | mAcc (%): 36.70 -> 42.12 | Helped |
| TransMatcher vs QAConv-GS under same code base | QAConv-GS · MSMT17 -> Market-1501 | Rank-1 (%): 75.1 -> 80.1 | Helped |
| add MLPHead1 (expand dimension to D) (ablation) | TransMatcher · MSMT17 ablation | mAcc (%): 41.44 -> 42.82 | Helped |
| add FC1 and BN1 together (ablation) | TransMatcher · MSMT17 ablation | mAcc (%): 42.82 -> 44.04 | Helped |
| add learnable prior score embedding (ablation) | TransMatcher · MSMT17 ablation | mAcc (%): 44.04 -> 44.29 | Helped |
| add learnable positional embedding in the encoders (ablation) | TransMatcher · MSMT17 ablation | mAcc (%): 44.04 -> 42.39 | Hurt |
| increase number of layers N beyond 3 (ablation) | TransMatcher · MSMT17 parameter analysis | mAcc (%) | No clear effect |
| increase number of heads H in encoders (ablation) | TransMatcher · MSMT17 parameter analysis | mAcc (%) | No clear effect |
| increase model dimension d (ablation) | TransMatcher · MSMT17 parameter analysis | mAcc (%) | Helped |
Closest papers by meaning
- Deep Neural Networks with Inexact Matching for Person Re-Identification (NeurIPS 2016 · similarity 0.89)
- TransforMatcher: Match-to-Match Attention for Semantic Correspondence (CVPR 2022 · similarity 0.87)
- NFormer: Robust Person Re-Identification With Neighbor Transformer (CVPR 2022 · similarity 0.87)
- Joint Learning of Single-Image and Cross-Image Representations for Person Re-Identification (CVPR 2016 · similarity 0.86)
- DC-Former: Diverse and Compact Transformer for Person Re-identification (AAAI 2023 · similarity 0.86)
- TokenMatcher: Diverse Tokens Matching for Unsupervised Visible-Infrared Person Re-Identification (AAAI 2025 · 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.