Research map / 3D vision / Face and shape
LCD: Learned Cross-Domain Descriptors for 2D-3D Matching
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the jointly trained 2D-3D descriptor instead of SuperPoint | SceneNN image matching · SceneNN image matching, Average | precision: 0.419 -> 0.422 | Helped |
| Train a shared cross-domain descriptor instead of a 2D-only auto-encoder (ablation) | SceneNN image matching · SceneNN image matching, Average | precision: 0.406 -> 0.422 | Helped |
| Reduce descriptor dimension from 256 to 128 (ablation) | LCD · SceneNN image matching, Average | precision: 0.422 -> 0.408 | Hurt |
| Reduce descriptor dimension from 256 to 64 (ablation) | LCD · SceneNN image matching, Average | precision: 0.422 -> 0.392 | Hurt |
| Use a shared cross-domain descriptor instead of a 3D-only auto-encoder (ablation) | 3DMatch registration · 3DMatch registration, Average | recall: 0.677 -> 0.725 | Helped |
| Reduce descriptor dimension from 256 to 128 (ablation) | LCD · 3DMatch registration, Average | recall: 0.725 -> 0.720 | Hurt |
| Reduce descriptor dimension from 256 to 64 (ablation) | LCD · 3DMatch registration, Average | recall: 0.725 -> 0.730 | Helped |
| Use LCD-D256 instead of 3DSmoothNet | 3DMatch registration · 3DMatch registration, Average | recall: 0.735 -> 0.725 | Hurt |
| Use RGB plus sparse depth with LCD-D64 instead of RGB-only FCRN | SceneNN depth estimation · SceneNN depth estimation | REL: 0.458 -> 0.187 | Helped |
| Use RGB plus sparse depth with LCD-D64 instead of RGB-only FCRN | SceneNN depth estimation · SceneNN depth estimation | δ3: 0.935 -> 0.908 | Hurt |
| Add Gaussian noise with standard deviation σ=0.1 to query images (ablation) | LCD 2D-3D place recognition · SceneNN 2D-3D place recognition | recall@10: ≈ 75% recall at 10 top candidates | Hurt |
Closest papers by meaning
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.88)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · similarity 0.85)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.85)
- Point2Real: Bridging the Gap between Point Cloud and Realistic Image for Open-World 3D Recognition (AAAI 2024 · similarity 0.85)
- CrossOver: 3D Scene Cross-Modal Alignment (CVPR 2025 · similarity 0.84)
- Uni3D: Exploring Unified 3D Representation at Scale (ICLR 2024 · similarity 0.84)
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.