LEST: Large-scale LiDAR Semantic Segmentation with Transformer
Not on the map. Why?
Abstract (arXiv)
Large-scale LiDAR-based point cloud semantic segmentation is a critical task in autonomous driving perception. Almost all of the previous state-of-the-art LiDAR semantic segmentation methods are variants of sparse 3D convolution. Although the Transformer architecture is becoming popular in the field of natural language processing and 2D computer vision, its application to large-scale point cloud semantic segmentation is still limited. In this paper, we propose a LiDAR sEmantic Segmentation architecture with pure Transformer, LEST. LEST comprises two novel components: a Space Filling Curve (SFC) Grouping strategy and a Distance-based Cosine Linear Transformer, DISCO. On the public nuScenes semantic segmentation validation set and SemanticKITTI test set, our model outperforms all the other state-of-the-art methods.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 33 of its 41 references that are in the map. First posted on arXiv 14 Jul 2023.
Builds on
33 of 41 references.
References · 33
- 2017Attention Is All You Needcited by 60,361
- 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
- 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,139
- 2016PointNet: Deep Learning on Point Sets for 3D Classification and Segmentationcited by 5,248
- 2017Squeeze-and-Excitation Networkscited by 4,789
- 2019nuScenes: A multimodal dataset for autonomous drivingcited by 4,523
- 2017PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Spacecited by 4,401
- 2020Transformers are RNNs: Fast Autoregressive Transformers with Linear Attentioncited by 2,022
- 2021Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutionscited by 1,962
- 2018PointPillars: Fast Encoders for Object Detection from Point Cloudscited by 1,715
- 2019KPConv: Flexible and Deformable Convolution for Point Cloudscited by 1,205
- 2020Center-based 3D Object Detection and Trackingcited by 1,144
21 more
- 2015Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)cited by 1,070
- 20173D Semantic Segmentation with Submanifold Sparse Convolutional Networkscited by 748
- 2020Long Range Arena: A Benchmark for Efficient Transformerscited by 700
- 2019RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Cloudscited by 585
- 2020Searching Efficient 3D Architectures with Sparse Point-Voxel Convolutioncited by 423
- 2021Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depthcited by 419
- 2020Cylindrical and Asymmetrical 3D Convolution Networks for LiDAR Segmentationcited by 389
- 2021DeepViT: Towards Deeper Vision Transformercited by 289
- 2020PolarNet: An Improved Grid Representation for Online LiDAR Point Clouds Semantic Segmentationcited by 275
- 2021Voxel Transformer for 3D Object Detectioncited by 247
- 2018SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloudcited by 228
- 2020SalsaNext: Fast, Uncertainty-aware Semantic Segmentation of LiDAR Point Clouds for Autonomous Drivingcited by 226
- 2020Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completioncited by 223
- 2022cosFormer: Rethinking Softmax in Attentioncited by 213
- 2021Embracing Single Stride 3D Object Detector with Sparse Transformercited by 193
- 2020SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentationcited by 170
- 2021Panoptic nuScenes: A Large-Scale Benchmark for LiDAR Panoptic Segmentation and Trackingcited by 166
- 2021SOFT: Softmax-free Transformer with Linear Complexitycited by 128
- 2020Low-Rank Bottleneck in Multi-head Attention Modelscited by 89
- 2021EPMF: Efficient Perception-aware Multi-sensor Fusion for 3D Semantic Segmentationcited by 59
- 2022MVP-Net: Multiple View Pointwise Semantic Segmentation of Large-Scale Point Cloudscited by 3
Built on by
Citing papers per year: 2024: 1.
1 paper.