Research map / Point clouds / Segmentation and detection
PillarNeXt: Rethinking Network Designs for 3D Object Detection in LiDAR Point Clouds
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 a pillar encoder instead of a voxel encoder at comparable FLOPs with enhanced training | PillarNeXt grid encoder · Waymo Open Dataset val, vehicle | 3D APH: 70.61 -> 70.64 | No clear effect |
| Use a pillar encoder instead of a voxel encoder at comparable FLOPs with enhanced training | PillarNeXt grid encoder · Waymo Open Dataset val, vehicle | BEV APH: 87.33 -> 88.13 | Helped |
| Use a pillar encoder instead of a voxel encoder | PillarNeXt-B · nuScenes val | NDS: 68.2 -> 68.8 | Helped |
| Replace the existing PillarNet neck with ASPP (ablation) | PillarNeXt · Waymo Open Dataset val, vehicle L2 | BEV AP: 84.54 -> 86.99 | Helped |
| Use a plain residual-block neck instead of the existing PillarNet neck (ablation) | PillarNeXt · Waymo Open Dataset val, vehicle L2 | BEV AP: 84.54 -> 83.86 | Hurt |
| Enlarge the receptive field with dilated blocks (ablation) | PillarNeXt neck · Waymo Open Dataset val, vehicle L2 | BEV AP: 84.54 -> 86.61 | Helped |
| Replace the PillarNet neck with FPN (ablation) | PillarNeXt · Waymo Open Dataset val, vehicle L2 | BEV AP: 84.54 -> 85.96 | Helped |
| Replace the PillarNet neck with BiFPN (ablation) | PillarNeXt · Waymo Open Dataset val, vehicle L2 | BEV AP: 84.54 -> 86.92 | Helped |
| Downsample output feature resolution from 0.3 to 0.6 (ablation) | PillarNeXt detection features · Waymo Open Dataset val, vehicle | 3D APH: 65.0 -> 62.8 | Hurt |
| Upsample features in the detection head (ablation) | PillarNeXt detection head · Waymo Open Dataset val, pedestrian | 3D APH: 66.6 -> 69.0 | Helped |
| Use a larger input grid size while holding output feature size at 0.3 (ablation) | PillarNeXt input grid · Waymo Open Dataset val, pedestrian | 3D APH: 69.0 -> 67.2 | Hurt |
Closest papers by meaning
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.92)
- GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (CVPR 2025 · similarity 0.89)
- Semi-supervised 3D Object Detection with PatchTeacher and PillarMix (AAAI 2024 · similarity 0.88)
- SwiftPillars: High-Efficiency Pillar Encoder for Lidar-Based 3D Detection (AAAI 2024 · similarity 0.88)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.88)
- Beyond Virtual Points: Depth-Enhanced LiDAR-only 3D Object Detection with Semi-Supervised Learning (Student Abstract) (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.