Research map / Graph learning / Graph algorithms and clustering
Deep Learning of Partial Graph Matching via Differentiable Top-K
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 |
|---|---|---|---|
| Add AFAT-U partial matching to GCAN | GCAN · Pascal VOC Keypoints (unfiltered) | F1: 59.7±0.3 -> 62.0±0.2 | Helped |
| Add AFAT-I partial matching | NGMv2 · Willow Object Class (+random outliers) | F1: 81.4±0.2 -> 83.1±0.2 | Helped |
| Add AFAT-I partial matching | NGMv2 · IMC-PT-SparseGM (50 anchors) | F1: 71.3±0.3 -> 72.8±0.4 | Helped |
| Add AFAT-U partial matching | GCAN · IMC-PT-SparseGM (100 anchors) | F1: 69.6±0.4 -> 71.5±0.2 | Helped |
| Train NGMv2 with differentiable top-k rather than adding top-k only at test time (ablation) | NGMv2 · Pascal VOC Keypoints (given ground-truth k) | F1: 60.7 -> 62.3 | Helped |
| Train NGMv2 with differentiable top-k rather than adding top-k only at test time (ablation) | NGMv2 · Willow Object Class (given ground-truth k) | F1: 85.2 -> 85.5 | No clear effect |
| Train NGMv2 with differentiable top-k rather than adding top-k only at test time (ablation) | NGMv2 · IMC-PT-SparseGM (50 anchors, given ground-truth k) | F1: 72.3 -> 73.6 | Helped |
| Train NGMv2 with differentiable top-k rather than adding top-k only at test time (ablation) | NGMv2 · IMC-PT-SparseGM (100 anchors, given ground-truth k) | F1: 70.7 -> 71.8 | Helped |
| Apply top-k selection with RRWM instead of a learned neural matching pipeline (ablation) | RRWM · Pascal VOC Keypoints (given ground-truth k) | F1: 20.3 -> 19.4 | Hurt |
| Apply top-k selection with RRWM instead of a learned neural matching pipeline (ablation) | RRWM · Willow Object Class (given ground-truth k) | F1: 20.1 -> 18.9 | Hurt |
| Apply top-k selection with RRWM instead of a learned neural matching pipeline (ablation) | RRWM · IMC-PT-SparseGM (50 anchors, given ground-truth k) | F1: 39.6 -> 39.1 | Hurt |
| Apply top-k selection with RRWM instead of a learned neural matching pipeline (ablation) | RRWM · IMC-PT-SparseGM (100 anchors, given ground-truth k) | F1: 34.6 -> 34.0 | Hurt |
Closest papers by meaning
- Learning Partial Graph Matching via Optimal Partial Transport (ICLR 2025 · similarity 0.91)
- Deep Learning of Graph Matching (CVPR 2018 · similarity 0.90)
- Deep Latent Graph Matching (ICML 2021 · similarity 0.90)
- IA-GM: A Deep Bidirectional Learning Method for Graph Matching (AAAI 2021 · similarity 0.90)
- M3C: A Framework towards Convergent, Flexible, and Unsupervised Learning of Mixture Graph Matching and Clustering (ICLR 2024 · similarity 0.88)
- Position: Graph Matching Systems Deserve Better Benchmarks (ICML 2025 · similarity 0.87)
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.