SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentation

By Chenfeng Xu, Bichen Wu, Zining Wang, Wei Zhan, Péter Vajda, Kurt Keutzer, Masayoshi Tomizuka. 2020

449 citations recorded by the source corpus.

LiDAR point-cloud segmentation is an important problem for many applications. For large-scale point cloud segmentation, the \textit{de facto} method is to project a 3D point cloud to get a 2D LiDAR image and use convolutions to process it. Despite the similarity between regular RGB and LiDAR images, we discover that the feature distribution of LiDAR images changes drastically at different image locations. Using standard convolutions to process such LiDAR images is problematic, as convolution filters pick up local features that are only active in specific regions in the image. As a result, the capacity of the network is under-utilized and the segmentation performance decreases. To fix this, we propose Spatially-Adaptive Convolution (SAC) to adopt different filters for different locations according to the input image. SAC can be computed efficiently since it can be implemented as a series of element-wise multiplications, im2col, and standard convolution. It is a general framework such that several previous methods can be seen as special cases of SAC. Using SAC, we build SqueezeSegV3 for LiDAR point-cloud segmentation and outperform all previous published methods by at least 3.7% mIoU on the SemanticKITTI benchmark with comparable inference speed.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Environmental Engineering, Efficiency & Compression

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Environmental Engineering

In Efficiency & Compression

Citations in the corpus

Visual Transformers: Token-based Image Representation and Processing for Computer Vision (2020, background): “Later work [11, 19, 20] interpret this “modulation” as a way to make convolution spatially adaptive and content-aware.”
FBNetV3: Joint Architecture-Recipe Search using Neural Acquisition Function (2020, background): “Other work downsamples input non-uniformly [40, 44, 26] to reduce computational cost.”
RPVNet: A Deep and Efficient Range-Point-Voxel Fusion Network for LiDAR Point Cloud Segmentation (2021, methodology): “A parallel track of works [36, 38, 23, 12] follows a spherical projection scheme, i.”
Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution (2020, unknown): “9 SqueezeSegV3-21 [68] 9.”
Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (2022, background): “Lately, point cloud analysis has emerged as a popular topic in 3D understanding, attracting attention from academia and industry (Qi et al., 2017a; Shi et al., 2019; Xu et al., 2020).”
(AF)2-S3Net: Attentive Feature Fusion with Adaptive Feature Selection for Sparse Semantic Segmentation Network (2021, methodology): “SqueezeSegV3 [39] stands on the shoulder of [36, 17], adopting a Spatially-Adaptive Convolution (SAC) to use different filters in different locations in relation to the input image.”
SalsaNext: Fast, Uncertainty-Aware Semantic Segmentation of LiDAR Point Clouds (2020, methodology): “Projection-based methods instead transform the 3D point cloud into various formats such as voxel cells [9], [10], [11], multi-view representation [12], lattice structure [13], [14], and rasterized images [1], [15], [16], [17].”