Point cloud classification: research map
1,097 papers on Point cloud classification in Point clouds, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 3 clusters and 10 approaches. The busiest year so far is 2025.
What it is
A point cloud is a set of 3D points, from LiDAR or depth sensors. Networks must handle unordered points of varying density; this topic covers architectures for recognising shapes and scenes, self-supervised pretraining, language-linked 3D understanding and scene motion.
Terms you will see on this page:
- Point cloud
- a set of 3D points describing a surface or scene.
- PointNet-style network
- a network that processes points individually and pools them, so order does not matter.
- Scene flow
- the 3D motion of each point between frames.
Start reading with the most cited papers here:
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · 9,885 citations)
- PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space (NeurIPS 2017 · 2,147 citations)
Within Point clouds, its share held steady from 50.9% in 2023–24 to 50.9% in 2025–26 (232 → 259 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Point cloud classification in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Self-supervised point cloud pretraining rising: 8% → 16% of the topic (19 → 42 papers)
- Point-language and open-vocabulary 3D rising: 15% → 23% of the topic (35 → 59 papers)
- LiDAR point cloud compression and completion rising: 8% → 12% of the topic (19 → 31 papers)
- Point cloud detection and tracking falling: 10% → 4% of the topic (24 → 11 papers)
- Local feature aggregation networks falling: 17% → 13% of the topic (39 → 34 papers)
Papers from 2025–2026 that are already cited most:
- Point Cloud Mamba: Point Cloud Learning via State Space Model (AAAI 2025 · 82 citations)
- Spiking Point Transformer for Point Cloud Classification (AAAI 2025 · 13 citations)
- Spectral Informed Mamba for Robust Point Cloud Processing (CVPR 2025 · 12 citations)
Approaches and key papers
Point cloud architectures · 489 papers
Keywords: adversarial, convolution, invariant
Approaches in this cluster:
- Local feature aggregation networks (257 papers)
Design point-set convolutions and feature extractors for point cloud recognition. - Equivariant point cloud networks (108 papers)
Build rotation-equivariant and invariant architectures for point clouds. - Efficient point sampling architectures (65 papers)
Improve point cloud backbones with sampling, scaling and spiking networks. - Adversarial point cloud attacks (59 papers)
Craft and defend against imperceptible adversarial perturbations on point clouds.
Most cited and most cited since 2024:
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · 9,885 citations)
- PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space (NeurIPS 2017 · 2,147 citations)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · 600 citations)
- HIMap: HybrId Representation Learning for End-to-end Vectorized HD Map Construction (CVPR 2024 · 29 citations)
Motion and dynamic point clouds · 319 papers
Keywords: motion, estimation, scene
Approaches in this cluster:
- Scene flow and dynamic point clouds (144 papers)
Estimate motion and process sequences of dynamic point clouds. - Point cloud detection and tracking (85 papers)
Apply point cloud networks to tracking, completion and anomaly detection tasks. - LiDAR point cloud compression and completion (90 papers)
Compress, upsample and complete LiDAR point clouds with learned models.
Most cited and most cited since 2024:
- GraspNet-1Billion: A Large-Scale Benchmark for General Object Grasping (CVPR 2020 · 667 citations)
- Point-NeRF: Point-Based Neural Radiance Fields (CVPR 2022 · 561 citations)
- RCooper: A Real-world Large-scale Dataset for Roadside Cooperative Perception (CVPR 2024 · 58 citations)
- SparseOcc: Rethinking Sparse Latent Representation for Vision-Based Semantic Occupancy Prediction (CVPR 2024 · 53 citations)
3D-language and point cloud pretraining · 289 papers
Keywords: language, pre, understanding
Approaches in this cluster:
- Point-language and open-vocabulary 3D (144 papers)
Align 3D point clouds with language for zero-shot and open-vocabulary understanding. - Self-supervised point cloud pretraining (113 papers)
Pretrain point representations via cross-modal, contrastive and diffusion objectives. - Masked point autoencoders (32 papers)
Pretrain point clouds by reconstructing masked patches, often multi-scale or rotation-invariant.
Most cited and most cited since 2024:
- Point-BERT: Pre-Training 3D Point Cloud Transformers With Masked Point Modeling (CVPR 2022 · 780 citations)
- PointCLIP: Point Cloud Understanding by CLIP (CVPR 2022 · 435 citations)
- LL3DA: Visual Interactive Instruction Tuning for Omni-3D Understanding Reasoning and Planning (CVPR 2024 · 84 citations)
- Point Cloud Mamba: Point Cloud Learning via State Space Model (AAAI 2025 · 82 citations)
