Research map / Point clouds / Point cloud classification
Unified and Efficient Point-Line Local Features
François Costa, Raphael Kreft, Felix Möller, Hardik Shah, Ramanathan Rajaraman, Eckhard Goedeke, Shaohui Liu, Rémi Pautrat, Marc Pollefeys
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 joint line training to the point network (ablation) | UPAL · MegaDepth relative pose | AUC @5deg: 55.2 -> 58.2 | Improved |
| remove angle field branch from DeepLSD, use gradient angle (ablation) | DeepLSD · HPatches line detection | repeatability / loc error: 27.2 -> 28.2 repeatability; 1.432 -> 1.341 loc err | Improved |
| compute gradient angle in parallel on GPU (ablation) | fast LSD · HPatches line detection | latency (ms): 36 -> 27 | Improved |
| downsample seed grid with stride 2 (ablation) | fast LSD · HPatches line detection | latency (ms): 27 -> 19 | Improved |
| keep only a subset of seed points (ablation) | fast LSD · HPatches line detection | latency (ms): 19 -> 17 | Improved |
| offload image preprocessing to GPU (ablation) | fast LSD · HPatches line detection | latency (ms): 17 -> 11 | Improved |
| add lines to points for visual localization | UPAL · 7Scenes Stairs | accuracy: 49.1 -> 54.6 | Improved |
| use UPAL point features vs ALIKED | UPAL · MegaDepth relative pose | AUC @5deg: 59.8 -> 58.2 | Hurt |
| use UPAL vs XFeat on indoor ScanNet | UPAL · ScanNet relative pose | AUC @20deg: 45.7 -> 41.4 | Hurt |
| unified network vs ALIKED + DeepLSD pipeline | UPAL · Runtime (RTX 2080 Ti) | GPU latency (ms): 286 -> 70 | Improved |
| UPAL vs DISK on ETH3D triangulation | UPAL · ETH3D triangulation | accuracy @1cm (%): 69.22 -> 64.80 | Hurt |
| UPAL vs DeepLSD on RDNIM homography estimation | UPAL · RDNIM | H estimation @5px: 60.5 -> 58.0 | Hurt |
Closest papers by meaning
- Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task (CVPR 2020 · similarity 0.88)
- ELSD: Efficient Line Segment Detector and Descriptor (ICCV 2021 · similarity 0.87)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.85)
- ScaleLSD: Scalable Deep Line Segment Detection Streamlined (CVPR 2025 · similarity 0.85)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.84)
- Towards Complementary Keypoint Detection via Mixture of Detectors (NeurIPS 2026 · 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.
