Research map / Label-efficient and robust learning / Recognition and long-tail
TopicFM: Robust and Interpretable Topic-Assisted Feature Matching
Khang Truong Giang, Soohwan Song, Sungho Jo
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 |
|---|---|---|---|
| Use 50 topics instead of 10 (ablation) | TopicFM · MegaDepth | relative pose estimation AUC (5°): 48.6 -> 49.1 | Improved |
| Increase topic count from 50 to 100 (ablation) | TopicFM · MegaDepth | relative pose estimation AUC (5°): 49.1 -> 49.0 | No clear effect |
| Increase covisible topics from 2 to 6 (ablation) | TopicFM · MegaDepth | relative pose estimation AUC (5°): 49.3 -> 52.7 | Improved |
| Increase covisible topics from 6 to 10 (ablation) | TopicFM · MegaDepth | relative pose estimation AUC (5°): 52.7 -> 54.1 | Improved |
| Increase covisible topics from 10 to 12 (ablation) | TopicFM · MegaDepth | relative pose estimation AUC (5°): 54.1 -> 53.7 | Hurt |
| Use topic-assisted matching rather than LoFTR | image matching · HPatches | homography estimation AUC (3px): 65.9 -> 67.3 | Improved |
| Use topic-assisted matching rather than MatchFormer | image matching · MegaDepth | relative pose estimation AUC (10°): 69.4 -> 70.1 | Improved |
| Use topic-assisted matching rather than LoFTR | image matching · ScanNet | relative pose estimation AUC (20°): 50.62 -> 50.91 | Improved |
| Use TopicFM rather than LoFTR | visual localization · InLoc | overall: 69.8 -> 70.9 | Improved |
| Use TopicFM rather than LoFTR | feature matching · 1200 × 896 runtime benchmark | runtime (ms): 500 -> 323 | Improved |
Later papers that compared with this method
2 later accepted papers report TopicFM as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- TextFM: Robust Semi-dense Feature Matching with Language Guidance (CVPR 2026 · better on 3 of 3 shared results)
- EDM: Efficient Deep Feature Matching (ICCV 2025 · 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
- Visual-Semantic Matching by Exploring High-Order Attention and Distraction (CVPR 2020 · similarity 0.87)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.86)
- Visual Semantic Reasoning for Image-Text Matching (ICCV 2019 · similarity 0.86)
- Matching Visual Features to Hierarchical Semantic Topics for Image Paragraph Captioning (IJCV 2022 · similarity 0.86)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.86)
- Composing Object Relations and Attributes for Image-Text Matching (CVPR 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.
