Research map / Point clouds / Segmentation and detection
LidarNAS: Unifying and Searching Neural Architectures for 3D Point Clouds
Chenxi Liu, Zhaoqi Leng, Pei Sun, Shuyang Cheng, Charles R. Qi, Yin Zhou, Mingxing Tan, Dragomir Anguelov
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 |
|---|---|---|---|
| evolve architecture with LidarNAS search from PointPillars-like | PointPillars-like · Waymo vehicle | L1 3D AP: 67.6 -> 73.2 | Improved |
| evolve architecture with LidarNAS search from RSN | RSN · Waymo vehicle | L1 3D AP: 75.2 -> 75.6 | Improved |
| evolve architecture with LidarNAS search from RSN | RSN · Waymo pedestrian | L1 3D AP: 77.1 -> 77.4 | Improved |
| add 2D U-Net on range image before voxelization to pillar view (ablation) | PointPillars-like · Waymo vehicle | L1 3D AP: 67.6 -> 72.3 | Improved |
| decrease voxel size from 0.32 to 0.25 (ablation) | LidarNASNet-P · Waymo vehicle | L1 3D AP: 72.3 -> 73.2 | Improved |
| widen RSN sparse U-Net channels from 64 to 91 instead of searching (ablation) | RSN · Waymo vehicle | latency (ms): 60.8 ms vs 49.3 ms | Hurt |
| add pillar view with sparse 2D U-Net in first stage (ablation) | RSN · Waymo vehicle | L1 BEV AP: 87.7 -> 88.6 | Improved |
| use warm starting instead of random sampling of architectures (ablation) | LidarNAS search · Waymo vehicle | AP and latency | Improved |
| use perspective view branch (ablation) | random architectures · Waymo | L1 3D AP | Hurt |
| use sparse branches on pillar view (ablation) | random architectures · Waymo | latency (ms): coefficient -0.03 vs dense 3.06 | Improved |
| use sparse branches on perspective view (ablation) | random architectures · Waymo | latency (ms): coefficient 6.90 vs dense -1.65 | Hurt |
| use anchor-free head with PointPillars backbone | PointPillars-like · Waymo vehicle | L1 3D AP: 63.3 -> 69.3 | Improved |
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.
- PointPillars — PointPillars: Fast Encoders for Object Detection From Point Clouds (CVPR 2019) · 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
- LiDAR R-CNN: An Efficient and Universal 3D Object Detector (CVPR 2021 · similarity 0.91)
- DOPS: Learning to Detect 3D Objects and Predict Their 3D Shapes (CVPR 2020 · similarity 0.90)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.89)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.88)
- LiDAR-Net: A Real-scanned 3D Point Cloud Dataset for Indoor Scenes (CVPR 2024 · similarity 0.88)
- A New Adversarial Perspective for LiDAR-based 3D Object Detection (AAAI 2025 · similarity 0.88)
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.
