Research map / 3D vision / 3D generation
AnyMatch: Supercharging Universal Multi-Modal Image Matching with Large-Scale Single-View Images
Meng Yang, Zizhuo Li, Linfeng Tang, Fan Fan, Jiayi Ma
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 |
|---|---|---|---|
| fine-tune matchers on AnyMatch data instead of MINIMA data | RoMa · Any-syn test (RGB-IR) | AUC@10deg pose: 60.62 -> 69.88 | Improved |
| fine-tune on AnyMatch data instead of MINIMA data | EDM · Any-syn test (RGB-Event) | AUC@10deg pose: 3.30 -> 9.32 | Improved |
| fine-tune on AnyMatch data instead of MINIMA data | EDM · METU-VisTIR (real RGB-IR) | AUC@10deg pose: 32.80 -> 35.80 | Improved |
| fine-tune on AnyMatch data instead of MINIMA data | LoFTR · METU-VisTIR (real RGB-IR) | AUC@10deg pose: 30.84 -> 25.83 | Hurt |
| fine-tune on AnyMatch data instead of MINIMA data | RoMa · DIODE (real RGB-Depth) | AUC@5px homography: 51.09 -> 49.83 | Hurt |
| fine-tune on AnyMatch data instead of MINIMA data | RoMa · MMIM Remote Sensing (zero-shot) | AUC@5px homography: 44.19 -> 44.06 | No clear effect |
| fine-tune on AnyMatch data instead of MINIMA data | EDM · MMIM Remote Sensing (zero-shot) | AUC@5px homography: 30.52 -> 38.69 | Improved |
| train on three modality pairs instead of RGB-IR only (ablation) | EDM + AnyMatch (no SGCV) · METU-VisTIR | AUC@10deg pose: 27.71 -> 31.24 | Improved |
| add sample-level geometric consistency verification (eta=0.6) (ablation) | EDM + AnyMatch · METU-VisTIR | AUC@10deg pose: 31.24 -> 35.80 | Improved |
| use a stricter SGCV threshold of 0.7 instead of 0.6 (ablation) | EDM + AnyMatch · METU-VisTIR | AUC@10deg pose: 35.80 -> 32.52 | Hurt |
| fine-tune pretrained matcher instead of training from scratch (ablation) | EDM + AnyMatch · METU-VisTIR | AUC@10deg pose: 14.16 -> 35.80 | Improved |
| 3D view transformation instead of 2D homography synthesis (ablation) | EDM + AnyMatch · METU-VisTIR | AUC@10deg pose: 30.78 -> 35.80 | Improved |
Closest papers by meaning
- Any2Point: Empowering Any-modality Transformers for Efficient 3D Understanding (ECCV 2024 · similarity 0.87)
- TAPA-MVS: Textureless-Aware PAtchMatch Multi-View Stereo (ICCV 2019 · similarity 0.86)
- Multiple Modalities Image Matching With Large-Scale Pre-Training (TPAMI 2026 · similarity 0.86)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.86)
- CoMatcher: Multi-View Collaborative Feature Matching (CVPR 2025 · similarity 0.85)
- WildCAT3D: Appearance-Aware Multi-View Diffusion in the Wild (NeurIPS 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.
