Research map / Probabilistic and scientific modeling / Molecules and proteins
DeepMatching: Hierarchical Deformable Dense Matching
Jérôme Revaud, Philippe Weinzaepfel, Zaïd Harchaoui, Cordelia Schmid
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 the new scoring (sum of correlations along backtracking path) (ablation) | DeepMatching R=1/2 · Mikolajczyk (viewpoint sequences) | accuracy@10: 0.848 -> 0.864 | Improved |
| add new backtracking entry points (all top-level positions) (ablation) | DeepMatching R=1/2 with new scoring · Mikolajczyk (viewpoint sequences) | accuracy@10: 0.864 -> 0.878 | Improved |
| add the new scoring (sum of correlations along backtracking path) (ablation) | DeepMatching R=1/2 · MPI-Sintel final (train) | accuracy@10: 0.880 -> 0.890 | Improved |
| add new backtracking entry points (all top-level positions) (ablation) | DeepMatching R=1/2 with new scoring · MPI-Sintel final (train) | accuracy@10: 0.890 -> 0.892 | Improved |
| add the new scoring (sum of correlations along backtracking path) (ablation) | DeepMatching R=1/2 · Kitti (train) | accuracy@10: 0.841 -> 0.855 | Improved |
| add new backtracking entry points (all top-level positions) (ablation) | DeepMatching R=1/2 with new scoring · Kitti (train) | accuracy@10: 0.855 -> 0.856 | Improved |
| increase input resolution from 1/4 to 1/2 | DeepMatching (preliminary) · Mikolajczyk (viewpoint sequences) | accuracy@10: 0.620 -> 0.848 | Improved |
| increase input resolution from 1/4 to 1/2 | DeepMatching (preliminary) · MPI-Sintel final (train) | accuracy@10: 0.822 -> 0.880 | Improved |
| use new backtracking entry points to cut cost (ablation) | DeepMatching R=1/2 with new scoring · Mikolajczyk (viewpoint sequences) | memory usage: 5.5 GB -> 4.4 GB | Improved |
| compress atomic patches with a 1024-prototype dictionary | DeepMatching R=1/2 · MPI-Sintel final (train) | accuracy@10: 0.892 -> 0.869 | Hurt |
| compress atomic patches with a 1024-prototype dictionary | DeepMatching R=1/2 · Kitti (train) | accuracy@10: 0.856 -> 0.812 | Hurt |
| compress atomic patches with a 1024-prototype dictionary | DeepMatching R=1/2 · Mikolajczyk (viewpoint sequences) | accuracy@10: 0.878 -> 0.820 | Hurt |
Closest papers by meaning
- Dense Flow from Adaptive Correspondence (NeurIPS 2026 · similarity 0.89)
- Deep Matching Prior: Test-Time Optimization for Dense Correspondence (ICCV 2021 · similarity 0.88)
- Flow Fields: Dense Correspondence Fields for Highly Accurate Large Displacement Optical Flow Estimation (TPAMI 2019 · similarity 0.88)
- Adapting Dense Matching for Homography Estimation with Grid-based Acceleration (CVPR 2025 · similarity 0.87)
- Diffusion Model for Dense Matching (ICLR 2024 · similarity 0.87)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 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.
