mlsanityData to 7 Oct 2026

PointConvFormer: Revenge of the Point-based Convolution

arXiv 2208.02879 ↗First posted 4 Aug 2022cs.CV cs.LGcited by 29

Not on the map. Why?

Abstract (arXiv)

We introduce PointConvFormer, a novel building block for point cloud based deep network architectures. Inspired by generalization theory, PointConvFormer combines ideas from point convolution, where filter weights are only based on relative position, and Transformers which utilize feature-based attention. In PointConvFormer, attention computed from feature difference between points in the neighborhood is used to modify the convolutional weights at each point. Hence, we preserved the invariances from point convolution, whereas attention helps to select relevant points in the neighborhood for convolution. PointConvFormer is suitable for multiple tasks that require details at the point level, such as segmentation and scene flow estimation tasks. We experiment on both tasks with multiple datasets including ScanNet, SemanticKitti, FlyingThings3D and KITTI. Our results show that PointConvFormer offers a better accuracy-speed tradeoff than classic convolutions, regular transformers, and voxelized sparse convolution approaches. Visualizations show that PointConvFormer performs similarly to convolution on flat areas, whereas the neighborhood selection effect is stronger on object boundaries, showing that it has got the best of both worlds.

Key facts. Cited by 29 papers in the map (data through 7 Oct 2026). Builds on 76 of its 107 references that are in the map. First posted on arXiv 4 Aug 2022.

Builds on

76 of 107 references.

References · 76

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  4. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  5. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  6. 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,223
  7. 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,384
  8. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,827
  9. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,088
  10. 2019XLNet: Generalized Autoregressive Pretraining for Language Understandingcited by 2,801
  11. 2018CBAM: Convolutional Block Attention Modulecited by 2,463
  12. 2017Non-local Neural Networkscited by 2,290
38 more
  1. 2019Transformer-XL: Attentive Language Models Beyond a Fixed-Length Contextcited by 1,830
  2. 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,707
  3. 2017Deformable Convolutional Networkscited by 1,481
  4. 2020Point Transformercited by 1,462
  5. 2019KPConv: Flexible and Deformable Convolution for Point Cloudscited by 1,202
  6. 20194D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networkscited by 1,201
  7. 2018Self-Attention with Relative Position Representationscited by 1,112
  8. 2015A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimationcited by 878
  9. 2018PointCNN: Convolution On X-Transformed Pointscited by 764
  10. 20173D Semantic Segmentation with Submanifold Sparse Convolutional Networkscited by 746
  11. 2016Multi-View 3D Object Detection Network for Autonomous Drivingcited by 721
  12. 2018PointConv: Deep Convolutional Networks on 3D Point Cloudscited by 656
  13. 2019RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Cloudscited by 584
  14. 2015Multi-view Convolutional Neural Networks for 3D Shape Recognitioncited by 564
  15. 2019Stand-Alone Self-Attention in Vision Modelscited by 542
  16. 2019DeepGCNs: Can GCNs Go as Deep as CNNs?cited by 527
  17. 2016Semantic Scene Completion from a Single Depth Imagecited by 468
  18. 2020Searching Efficient 3D Architectures with Sparse Point-Voxel Convolutioncited by 422
  19. 2021Conditional Positional Encodings for Vision Transformerscited by 402
  20. 2020Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentationcited by 388
  21. 2019Dynamic Convolution: Attention over Convolution Kernelscited by 355
  22. 2016OctNet: Learning Deep 3D Representations at High Resolutionscited by 340
  23. 2016Dynamic Filter Networkscited by 318
  24. 2020Conditional Convolutions for Instance Segmentationcited by 314
  25. 2020Exploring Self-attention for Image Recognitioncited by 313
  26. 2017Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphscited by 307
  27. 2019CondConv: Conditionally Parameterized Convolutions for Efficient Inferencecited by 273
  28. 2018SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filterscited by 261
  29. 2017Learning SO(3) Equivariant Representations with Spherical CNNscited by 244
  30. 2018FlowNet3D: Learning Scene Flow in 3D Point Cloudscited by 241
  31. 2020PointASNL: Robust Point Clouds Processing using Nonlocal Neural Networks with Adaptive Samplingcited by 238
  32. 2019Pay Less Attention with Lightweight and Dynamic Convolutionscited by 211
  33. 2019Local Relation Networks for Image Recognitioncited by 210
  34. 2020SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentationcited by 170
  35. 2022Stratified Transformer for 3D Point Cloud Segmentationcited by 166
  36. 2021(AF)2-S3Net: Attentive Feature Fusion with Adaptive Feature Selection for Sparse Semantic Segmentation Networkcited by 164
  37. 2021Deep Parametric Continuous Convolutional Neural Networkscited by 161
  38. 2018SPLATNet: Sparse Lattice Networks for Point Cloud Processingcited by 158

50 most cited of 76 All 76 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 10; 2024: 10; 2025: 6; 2026: 3.

29 papers.

Sort:

By area:

  1. 2023Point Transformer V3: Simpler, Faster, Strongercited by 579
  2. 2023Swin3D: A Pretrained Transformer Backbone for 3D Indoor Scene Understandingcited by 111
  3. 2025Sonata: Self-Supervised Learning of Reliable Point Representationscited by 108
  4. 2023Towards Large-scale 3D Representation Learning with Multi-dataset Point Prompt Trainingcited by 85
  5. 2023Mask-Attention-Free Transformer for 3D Instance Segmentationcited by 45
  6. 2025Grounding 3D Object Affordance with Language Instructions, Visual Observations and Interactionscited by 19
  7. 2025LitePT: Lighter Yet Stronger Point Transformercited by 17
  8. 2024KPConvX: Modernizing Kernel Point Convolution with Kernel Attentioncited by 16
  9. 20243D Open-Vocabulary Panoptic Segmentation with 2D-3D Vision-Language Distillationcited by 13
  10. 2023Multi-Body Neural Scene Flowcited by 12
  11. 2024DiffSF: Diffusion Models for Scene Flow Estimationcited by 6
  12. 2023Spherical Frustum Sparse Convolution Network for LiDAR Point Cloud Semantic Segmentationcited by 5
17 more
  1. 2024FlowMamba: Learning Point Cloud Scene Flow with Global Motion Propagationcited by 5
  2. 2024Advancements in Point Cloud-Based 3D Defect Detection and Classification for Industrial Systems: A Comprehensive Surveycited by 3
  3. 2024Object Dynamics Modeling with Hierarchical Point Cloud-based Representationscited by 3
  4. 2023Self-Supervised 3D Scene Flow Estimation and Motion Prediction using Local Rigidity Priorcited by 2
  5. 2024PoCo: Point Context Cluster for RGBD Indoor Place Recognitioncited by 2
  6. 2024BiEquiFormer: Bi-Equivariant Representations for Global Point Cloud Registrationcited by 1
  7. 2024CSCPR: Cross-Source-Context Indoor RGB-D Place Recognitioncited by 1
  8. 2025SAMBLE: Shape-Specific Point Cloud Sampling for an Optimal Trade-Off Between Local Detail and Global Uniformitycited by 1
  9. 2025HyperPointFormer: Multimodal Fusion in 3D Space with Dual-Branch Cross-Attention Transformerscited by 1
  10. 2025Enhancing point cloud analysis via neighbor aggregation correction based on cross-stage structure correlationcited by 1
  11. 2026RelFlexformer: Efficient Attention 3D-Transformers for Integrable Relative Positional Encodingscited by 1
  12. 2023Long Range Pooling for 3D Large-Scale Scene Understandingcited by 0
  13. 2023Point Neighborhood Embeddingscited by 0
  14. 2023Bidirectional Knowledge Reconfiguration for Lightweight Point Cloud Analysiscited by 0
  15. 2024Long-Term 3D Point Tracking By Cost Volume Fusioncited by 0
  16. 2026PCSTracker: Long-Term Scene Flow Estimation for Point Cloud Sequencescited by 0
  17. 2026Channel-Level Relation to Attentive Aggregation with Neighborhood-Homogeneity Constraint for Point Cloud Analysiscited by 0