Research map / Point clouds / Registration and matching
Learnable Graph Matching: A Practical Paradigm for Data Association
Jiawei He, Zehao Huang, Naiyan Wang, Zhaoxiang Zhang
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 |
|---|---|---|---|
| add graph matching layer at inference only (no training) (ablation) | DeepSORT baseline · MOT17 val (3-fold CV) | IDF1: 69.5 -> 70.2 | Improved |
| train with the graph matching layer in addition to inference (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 70.2 -> 71.5 | Improved |
| add graph matching (QP layer) over Hungarian-based DeepSORT (ablation) | DeepSORT · MOT17 val (3-fold CV) | IDF1: 68.1 -> 70.0 | Improved |
| add geometric information in cross-graph GCN (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 70.6 -> 71.5 | Improved |
| add appearance encoder and GCN (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 70.0 -> 70.6 | Improved |
| add linear interpolation post-processing (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 71.5 -> 73.0 | Improved |
| aggregate tracklet features by mean instead of last frame (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 64.3 -> 71.5 | Improved |
| use moving average instead of mean aggregation (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 71.5 vs 69.8-70.6 | No clear effect |
| add edge GCN on top of vertex GCN (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 71.8 -> 72.0 | No clear effect |
| add residual connections to 2-layer GCN (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 71.1 -> 71.8 | Improved |
| increase max age from 30 to 100 frames (ablation) | GMTracker · MOT17 val (3-fold CV) | IDF1: 69.2 -> 71.5 | Improved |
| increase max age from 100 to 150 frames (ablation) | DeepSORT · MOT17 val (3-fold CV) | IDF1: 68.1 -> 67.3 | Hurt |
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.
- GeoTransformer — GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer (TPAMI 2023) · better on 1 of 1 shared result
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
- Learnable Graph Matching: Incorporating Graph Partitioning With Deep Feature Learning for Multiple Object Tracking (CVPR 2021 · similarity 0.91)
- Deep Learning of Graph Matching (CVPR 2018 · similarity 0.89)
- Ensemble Quadratic Assignment Network for Graph Matching (IJCV 2024 · similarity 0.88)
- Neural Graph Matching Network: Learning Lawler’s Quadratic Assignment Problem with Extension to Hypergraph and Multiple-graph Matching (TPAMI 2021 · similarity 0.88)
- Discrete Cycle-Consistency Based Unsupervised Deep Graph Matching (AAAI 2024 · similarity 0.87)
- Learning Partial Graph Matching via Optimal Partial Transport (ICLR 2025 · similarity 0.86)
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.
