mlsanityData to 14 Oct 2026

Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality

arXiv 2412.16481 ↗First posted 21 Dec 2024cs.CVcited by 1

Not on the map. Why?

Abstract (arXiv)

Recent efforts recognize the power of scale in 3D learning (e.g. PTv3) and attention mechanisms (e.g. FlashAttention). However, current point cloud backbones fail to holistically unify geometric locality, attention mechanisms, and GPU architectures in one view. In this paper, we introduce Flash3D Transformer, which aligns geometric locality and GPU tiling through a principled locality mechanism based on Perfect Spatial Hashing (PSH). The common alignment with GPU tiling naturally fuses our PSH locality mechanism with FlashAttention at negligible extra cost. This mechanism affords flexible design choices throughout the backbone that result in superior downstream task results. Flash3D outperforms state-of-the-art PTv3 results on benchmark datasets, delivering a 2.25x speed increase and 2.4x memory efficiency boost. This efficiency enables scaling to wider attention scopes and larger models without additional overhead. Such scaling allows Flash3D to achieve even higher task accuracies than PTv3 under the same compute budget.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 32 of its 48 references that are in the map. First posted on arXiv 21 Dec 2024.

Builds on

32 of 48 references.

References · 32

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  2. 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,139
  3. 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,523
  4. 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
  5. 2023FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioningcited by 2,598
  6. 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,451
  7. 2020Point Transformercited by 1,470
  8. 20194D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networkscited by 1,208
  9. 2022Planning-oriented Autonomous Drivingcited by 1,087
  10. 2023Point Transformer V3: Simpler, Faster, Strongercited by 589
  11. 2024FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precisioncited by 519
  12. 2020Searching Efficient 3D Architectures with Sparse Point-Voxel Convolutioncited by 423
20 more
  1. 2022Point Transformer V2: Grouped Vector Attention and Partition-based Poolingcited by 388
  2. 2018Efficient Attention: Attention with Linear Complexitiescited by 368
  3. 2021Embracing Single Stride 3D Object Detector with Sparse Transformercited by 193
  4. 2021Panoptic nuScenes: A Large-Scale Benchmark for LiDAR Panoptic Segmentation and Trackingcited by 166
  5. 2021(AF)2-S3Net: Attentive Feature Fusion with Adaptive Feature Selection for Sparse Semantic Segmentation Networkcited by 164
  6. 2018Online normalizer calculation for softmaxcited by 162
  7. 2023Spherical Transformer for LiDAR-based 3D Recognitioncited by 125
  8. 2023Swin3D: A Pretrained Transformer Backbone for 3D Indoor Scene Understandingcited by 113
  9. 2020Cylinder3D: An Effective 3D Framework for Driving-scene LiDAR Semantic Segmentationcited by 112
  10. 2023OctFormer: Octree-based Transformers for 3D Point Cloudscited by 109
  11. 2022SWFormer: Sparse Window Transformer for 3D Object Detection in Point Cloudscited by 92
  12. 2022DexPoint: Generalizable Point Cloud Reinforcement Learning for Sim-to-Real Dexterous Manipulationcited by 84
  13. 2023FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformercited by 73
  14. 2021Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusioncited by 72
  15. 2023FusionAD: Multi-modality Fusion for Prediction and Planning Tasks of Autonomous Drivingcited by 67
  16. 2024ThunderKittens: Simple, Fast, and Adorable AI Kernelscited by 36
  17. 2021BuildingNet: Learning to Label 3D Buildingscited by 33
  18. 2019Data-Efficient Learning for Sim-to-Real Robotic Grasping using Deep Point Cloud Prediction Networkscited by 6
  19. 2020Point Cloud Based Reinforcement Learning for Sim-to-Real and Partial Observability in Visual Navigationcited by 5
  20. 2023Toward Zero-Shot Sim-to-Real Transfer Learning for Pneumatic Soft Robot 3D Proprioceptive Sensingcited by 1

Built on by

Cited by year

Citing papers per year: 2025: 1.

1 paper.

  1. 2025LitePT: Lighter Yet Stronger Point Transformercited by 17