Research map / Point clouds / Segmentation and detection
LiDAR voxel and pillar detectors
Detect 3D objects in driving LiDAR with efficient voxel, pillar and domain-adaptive designs.
56 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Segmentation and detection, keyword cluster “detection · 3d object · detectors”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Few recent reports The tracked benchmarks have at most one report from 2025–26 that could beat the old record, too few to say whether they are still moving.
20 papers in 2023–24 and 15 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 20.0% to 16.5%.
Open questions the data points to
- Papers report PointPillars on nuScenes test (mAP) anywhere from 30.5 to 40.1: comparisons against it depend on whose number you take.
- Papers report PointPillars on nuScenes test (NDS) anywhere from 45.3 to 55: comparisons against it depend on whose number you take.
No longer reported by papers here: nuScenes test (NDS, last in 2024); nuScenes val (NDS, last in 2024); nuScenes val (mAP, last in 2024).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Multi-View 3D Object Detection Network for Autonomous Driving (CVPR 2017 · 3,503 citations)
- 2020 · PointPainting: Sequential Fusion for 3D Object Detection (CVPR 2020 · 1,229 citations)
- 2021 · PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (CVPR 2021 · 472 citations)
- 2022 · Not All Points Are Equal: Learning Highly Efficient Point-Based Detectors for 3D LiDAR Point Clouds (CVPR 2022 · 413 citations)
- 2023 · Spherical Transformer for LiDAR-Based 3D Recognition (CVPR 2023 · 239 citations)
- 2024 · PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · 19 citations)
- 2025 · UniMamba: Unified Spatial-Channel Representation Learning with Group-Efficient Mamba for LiDAR-based 3D Object Detection (CVPR 2025 · 17 citations)
- 2026 · Physically-Based LiDAR Smoke Simulation for Robust 3D Object Detection (AAAI 2026 · 1 citation)
Earliest papers here
- Multi-View 3D Object Detection Network for Autonomous Driving (CVPR 2017 · 3,503 citations)
- PointPainting: Sequential Fusion for 3D Object Detection (CVPR 2020 · 1,229 citations)
- Learning Depth-Guided Convolutions for Monocular 3D Object Detection (CVPR 2020 · 275 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
nuScenes test · mAP (higher is better)
10 papers, 2020–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 46.4 | PointPainting: Sequential Fusion for 3D Object Detection (CVPR 2020 · 1,229 citations) |
| 2021 | 66.8 | PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (CVPR 2021 · 472 citations) |
| 2022 | 69.2 | BEVFusion: A Simple and Robust LiDAR-Camera Fusion Framework (NeurIPS 2022 · 201 citations) |
| 2024 | 69.6 | BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · 1 citation) |
| 2025 | 70.2 | UniMamba: Unified Spatial-Channel Representation Learning with Group-Efficient Mamba for LiDAR-based 3D Object Detection (CVPR 2025 · 17 citations) |
nuScenes test · NDS (higher is better)
9 papers, 2020–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 58.1 | PointPainting: Sequential Fusion for 3D Object Detection (CVPR 2020 · 1,229 citations) |
| 2021 | 71 | PointAugmenting: Cross-Modal Augmentation for 3D Object Detection (CVPR 2021 · 472 citations) |
| 2022 | 71.8 | BEVFusion: A Simple and Robust LiDAR-Camera Fusion Framework (NeurIPS 2022 · 201 citations) |
| 2023 | 72.7 | MGTANet: Encoding Sequential LiDAR Points Using Long Short-Term Motion-Guided Temporal Attention for 3D Object Detection (AAAI 2023 · 30 citations) |
| 2024 | 73.6 | BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · 1 citation) |
nuScenes val · NDS (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 70.8 | Multimodal Virtual Point 3D Detection (NeurIPS 2021 · 125 citations) |
nuScenes val · mAP (higher is better)
4 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 67.1 | Multimodal Virtual Point 3D Detection (NeurIPS 2021 · 125 citations) |
What has been tried
49 papers here report 554 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 6%, hurt 16%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add both long-term and short-term memory encoders | Waymo Open Dataset val, L1 and L2 · mAP/mAPH | 79.61/78.22 -> 81.38/80.04; 74.21/72.79 -> 75.88/74.59 | Helped | PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · 19 citations) |
| Add spatial locality modeling (SLM) | nuScenes val · mAP | 67.6 -> 67.9 | Helped | UniMamba: Unified Spatial-Channel Representation Learning with Group-Efficient Mamba for LiDAR-based 3D Object Detection (CVPR 2025 · 17 citations) |
| combine voxel and pillar representations | Waymo Open Dataset val, LEVEL 2 · mAPH | 67.5/66.3 -> 68.3 | Helped | Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · 10 citations) |
| Add DCG to the vanilla model | KITTI val, Car Moderate (IoU=0.7) · AP | 82.08 -> 88.86 | Helped | iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · 7 citations) |
| Reconstruct only the neighborhood of visible voxels | nuScenes val, NonLP · mIoU | 59.0 -> 66.7 | Helped | Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in LiDAR Point Clouds (CVPR 2025 · 5 citations) |
| Use two sides fusion instead of body side, head/tail side, or corner point fusion | KITTI, TED, water mist · attack success rate | 61.01, 55.30, 57.22 -> 86.98 | Helped | A New Adversarial Perspective for LiDAR-based 3D Object Detection (AAAI 2025 · 3 citations) |
| Add the ray generator | S-KITTI (all weather) · 3DR40 AP (%) | 80.29 -> 81.71 | Helped | Equirectangular Point Reconstruction for Domain Adaptive Multimodal 3D Object Detection in Adverse Weather Conditions (AAAI 2025 · 2 citations) |
| Remove CCE loss from REAL | SemanticKITTI · AUPR | 20.00 -> 26.68 | Helped | LiON: Learning Point-Wise Abstaining Penalty for LiDAR Outlier DetectioN Using Diverse Synthetic Data (AAAI 2025 · 2 citations) |
| Add local mean feature to global coordinates | nuScenes→KITTI · mAP 3D | 21.07 -> 22.45 | Helped | GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (CVPR 2025 · 2 citations) |
| Add synthetic data without JiSAM components | NuScenes · mAP | 60.18 -> 58.29 (-1.89) | Hurt | JiSAM: Alleviate Labeling Burden and Corner Case Problems in Autonomous Driving via Minimal Real-World Data (CVPR 2025 · 2 citations) |
| Add height filtering before uniform ray sampling | NuScenes · mAP | 31.68 -> 33.17 | Helped | TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · 1 citation) |
| Add shared learnable point token | Waymo, 20% labeled data · L2 mAP | 66.49 -> 66.70 | Helped | BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · 1 citation) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| PointPillars | nuScenes test | mAP | 30.5 – 40.1 | 3 papers, e.g. CVPR 2021 and CVPR 2020 |
| PointPillars | nuScenes test | NDS | 45.3 – 55 | 3 papers, e.g. CVPR 2021 and CVPR 2020 |
What reviewers will expect
Datasets and baselines named in the 9 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- nuScenes · 7 of 9 papers
- KITTI · 3 of 9 papers
- SemanticKITTI · 2 of 9 papers
- Waymo · 2 of 9 papers
- ONCE · 2 of 9 papers
- Waymo Open Dataset · 2 of 9 papers
Baselines
- VoxelNeXt · 4 of 9 papers
- CenterPoint · 4 of 9 papers
- TED · 2 of 9 papers
- PV-RCNN · 2 of 9 papers
- TransFusion-L · 2 of 9 papers
- VirConv-S · 2 of 9 papers
- SFD · 2 of 9 papers
- GD-MAE · 2 of 9 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 35 papers; from OpenAlex and the papers' affiliations).
- The Chinese University of Hong Kong · 4 papers
- Tsinghua University · 2 papers
- University of Macau · 2 papers
- Beijing Institute of Technology · 2 papers
- University of California, Merced · 2 papers
- Xiamen University · 2 papers
- Korea University · 2 papers
- QCraft · 2 papers
- The University of Hong Kong · 2 papers
- City University of Hong Kong · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Single-stage point-voxel detection (45 papers)
- Multimodal and open-vocabulary 3D detection (24 papers)
Elsewhere in the topic
- Point cloud semantic segmentation (108 papers)
- LiDAR segmentation for driving (47 papers)
- Scene completion and instance segmentation (44 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).