Research map / Point clouds / Point cloud classification
C-LaV: Conditional Latent Velocity Field Denoising for Weather-Robust LiDAR Place Recognition
Xuewei Cao, Jiayue Yang, Zhiwen Zeng, Yanyong Zhang, Yan Xia
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 |
|---|---|---|---|
| upgrade BEV encoder from DINOv2-S to DINOv2-B (ablation) | C-LaV with DDPM denoiser and NetVLAD · KITTI average | Recall@1 (%): 11.17% -> 30.45% | Improved |
| upgrade BEV encoder from DINOv2-S to DINOv2-B (ablation) | C-LaV with DDPM denoiser and NetVLAD · NCLT average | Recall@1 (%): 5.40% -> 16.80% | Improved |
| replace DDPM latent denoiser with flow-matched conditional velocity field and ODE sampler (ablation) | C-LaV (DINOv2-B, NetVLAD) · KITTI average | Recall@1 (%): 30.45% -> 50.15% | Improved |
| replace DDPM latent denoiser with flow-matched velocity field (ablation) | C-LaV (DINOv2-B, NetVLAD) · NCLT average | Recall@1 (%): 16.80% -> 27.35% | Improved |
| replace NetVLAD with SALAD latent clustering descriptor head (ablation) | C-LaV · KITTI average | Recall@1 (%): 50.15% -> 62.83% | Improved |
| replace NetVLAD with SALAD latent clustering descriptor head (ablation) | C-LaV · NCLT average | Recall@1 (%): 27.35% -> 34.52% | Improved |
| add BEV-space U-Net denoising before frozen DINOv2 and SALAD (ablation) | denoise-then-retrieve baseline · KITTI average | Recall@1 (%): 29.32% -> 37.00% | Improved |
| denoise in latent space with flow matching instead of BEV signal space (ablation) | denoise-then-retrieve baseline · KITTI average | Recall@1 (%): 37.00% -> 62.84% | Improved |
| C-LaV versus MinkLoc3D v2 under fog | C-LaV · KITTI fog | Recall@1 (%): 67.07 -> 62.73 | Hurt |
| C-LaV versus MinkLoc3D v2 under fog | C-LaV · NCLT fog | Recall@1 (%): 29.81 -> 28.87 | Hurt |
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.
- BEVPlace++ — BEVPlace: Learning LiDAR-based Place Recognition using Bird's Eye View Images (ICCV 2023) · better on 2 of 2 shared results
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
- No Thing, Nothing: Highlighting Safety-Critical Classes for Robust LiDAR Semantic Segmentation in Adverse Weather (CVPR 2025 · similarity 0.86)
- UniMix: Towards Domain Adaptive and Generalizable LiDAR Semantic Segmentation in Adverse Weather (CVPR 2024 · similarity 0.85)
- STGC-NeRF: Spatial-Temporal Geometric Consistency for LiDAR Neural Radiance Fields in Dynamic Scenes (AAAI 2025 · similarity 0.85)
- L2RSI: Cross-view LiDAR-based Place Recognition for Large-scale Urban Scenes via Remote Sensing Imagery (NeurIPS 2025 · similarity 0.85)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.85)
- WeatherGen: A Unified Diverse Weather Generator for LiDAR Point Clouds via Spider Mamba Diffusion (CVPR 2025 · similarity 0.85)
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.
