Research map / 3D vision / Depth and stereo
Hierarchical Metric Learning and Matching for 2D and 3D Geometric Correspondences
Mohammed E. Fathy, Quoc-Huy Tran, M. Zeeshan Zia, Paul Vernaza, Manmohan Chandraker
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 shallow conv2 features instead of deep conv5 features at small threshold | single-layer net · KITTI Flow 2015 PCK@5px | PCK: 61.78% -> 73.89% | Improved |
| use deep conv5 features instead of conv2 at large threshold | single-layer net · KITTI Flow 2015 PCK@15px | PCK: 81.36% -> 87.57% | Improved |
| add hypercolumn fusion over conv5-net | conv5-net · KITTI Flow 2015 PCK@5px | PCK: 61.78% -> 69.41% | Improved |
| hypercolumn fusion versus plain conv2 features | conv2-net · KITTI Flow 2015 PCK@5px | PCK: 73.89% -> 69.41% | Hurt |
| add hierarchical matching with deep-to-shallow refinement | conv2-net · KITTI Flow 2015 PCK@5px | PCK: 73.89% -> 80.17% | Improved |
| train on MPI Sintel instead of KITTI (cross-domain) (ablation) | HiLM (conv2+conv5) · KITTI Flow 2015 PCK@5px | PCK: 79.11% -> 72.37% | Hurt |
| HiLM trained on Sintel versus fusion baseline trained on KITTI | HiLM (conv2+conv5) · KITTI Flow 2015 PCK@5px | PCK: 69.41% -> 72.37% | Improved |
| add hierarchical metric learning and matching versus UCN | UCN · KITTI Flow 2015 PCK@2px | PCK: 29.38% -> 43.26% | Improved |
| switch backbone from VGG-M to GoogLeNet (ablation) | HiLM · KITTI Flow 2015 PCK@10px | PCK: 85.08% -> 85.18% | No clear effect |
| add deep supervised shallow features to HiL(conv8) (ablation) | 3D CNN · 3DMatch PCK@9cm | PCK: 20.78% -> 21.50% | Improved |
| add shallow refinement on top of deep features (HiLM) | 3DMatch · 3DMatch PCK@9cm | PCK: 22.04% -> 24.36% | Improved |
| use HiLM matches with EpicFlow interpolation versus semantic-prior methods | HiLM · KITTI Flow 2015 Fl-fg | outlier %: 8.75% vs 21.79% | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 3DMatch — 3DMatch: Learning Local Geometric Descriptors From RGB-D Reconstructions (CVPR 2017) · 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
- Convolutional Neural Network Architecture for Geometric Matching (CVPR 2017 · similarity 0.91)
- Learning 3D Keypoint Descriptors for Non-Rigid Shape Matching (ECCV 2018 · similarity 0.91)
- Deep Shape Matching (ECCV 2018 · similarity 0.90)
- Universal Correspondence Network (NeurIPS 2016 · similarity 0.89)
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.87)
- MATCHA: Towards Matching Anything (CVPR 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.
