Research map / Label-efficient and robust learning / Test-time adaptation
Deep Matching Prior: Test-Time Optimization for Dense Correspondence
Sunghwan Hong, Seungryong Kim
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 residual correspondence module (soft-argmax residual) (ablation) | DMP baseline · HPatches AEE | AEE: 33.1 -> 18.6 | Improved |
| add adaptation feature layer alone (ablation) | DMP baseline · HPatches AEE | AEE: 33.1 -> 32.8 | No clear effect |
| add adaptation feature layer on top of residual correspondence (ablation) | DMP · HPatches AEE | AEE: 18.6 -> 12.3 | Improved |
| add confidence threshold to contrastive loss (ablation) | DMP · HPatches AEE | AEE: 12.3 -> 9.69 | Improved |
| increase number of loss samples M from 128 to 256 or 512 (ablation) | DMP · HPatches AEE | AEE: 9.84 -> 9.69 -> 9.67 | No clear effect |
| set confidence threshold too high (ablation) | DMP · HPatches | AEE | Hurt |
| add test-time augmentation (A-DMP) | DMP · HPatches (240x240) AEE | AEE: 9.69 -> 8.18 | Improved |
| add test-time augmentation (A-DMP) | DMP · HPatches original resolution AEE | AEE: 30.64 -> 27.22 | Improved |
| add RANSAC coarse alignment (RANSAC-DMP) | DMP · HPatches (240x240) AEE | AEE: 9.69 -> 3.05 | Improved |
| add pre-training and test-time optimization (DMP†) | DMP · HPatches (240x240) AEE | AEE: 9.69 -> 5.21 | Improved |
| add test-time optimization to pre-trained DMP† | DMP† · HPatches (240x240) AEE | AEE: 7.53 -> 5.21 | Improved |
| add test-time optimization to pre-trained RANSAC-Flow | RANSAC-Flow · HPatches (240x240) AEE | AEE: 3.06 -> 5.19 | Hurt |
Closest papers by meaning
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.92)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.90)
- COTR: Correspondence Transformer for Matching Across Images (ICCV 2021 · similarity 0.90)
- Towards Geometry-Grounded Dense Semantic Matching with VGGT Priors (ECCV 2026 · similarity 0.89)
- RoMa v2: Harder Better Faster Denser Feature Matching (ECCV 2026 · similarity 0.89)
- Learning Contrastive Representation for Semantic Correspondence (IJCV 2022 · similarity 0.89)
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.
