SqueezeSegV3: Spatially-Adaptive Convolution for Efficient Point-Cloud Segmentation (2020-04-03T00:00:00.000000Z)

TL;DR

Spatially-Adaptive Convolution (SAC) is proposed to adopt different filters for different locations according to the input image to improve LiDAR point-cloud segmentation and outperform all previous published methods by at least 3.7% mIoU on the SemanticKITTI benchmark with comparable inference speed.

Abstract

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.

Authors

K. Keutzer

18 papers

Bichen Wu

6 papers

Chenfeng Xu

2 papers

TL;DR

Abstract

Authors

References66 items

xMUDA: Cross-Modal Unsupervised Domain Adaptation for 3D Semantic Segmentation

RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds

RangeNet ++: Fast and Accurate LiDAR Semantic Segmentation

SuMa++: Efficient LiDAR-based Semantic SLAM

Deformable Kernel Networks for Joint Image Filtering

End-to-End Multi-View Fusion for 3D Object Detection in LiDAR Point Clouds

3D Neighborhood Convolution: Learning Depth-Aware Features for RGB-D and RGB Semantic Segmentation

A Review of Point Cloud Semantic Segmentation

Efficient Deep Neural Networks

Learn to Scale: Generating Multipolar Normalized Density Maps for Crowd Counting

Point-Voxel CNN for Efficient 3D Deep Learning

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

Searching for MobileNetV3

LATTE: Accelerating LiDAR Point Cloud Annotation via Sensor Fusion, One-Click Annotation, and Tracking

Pixel-Adaptive Convolutional Neural Networks

SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences

ChamNet: Towards Efficient Network Design Through Platform-Aware Model Adaptation

FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

PartNet: A Large-Scale Benchmark for Fine-Grained and Hierarchical Part-Level 3D Object Understanding

Learning to Sample

Deformable ConvNets V2: More Deformable, Better Results

Synetgy: Algorithm-hardware Co-design for ConvNet Accelerators on Embedded FPGAs

VV-Net: Voxel VAE Net With Group Convolutions for Point Cloud Segmentation

Gather-Excite: Exploiting Feature Context in Convolutional Neural Networks

Mixed Precision Quantization of ConvNets via Differentiable Neural Architecture Search

SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud

Fully-Convolutional Point Networks for Large-Scale Point Clouds

MnasNet: Platform-Aware Neural Architecture Search for Mobile

ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design

CBAM: Convolutional Block Attention Module

Efficient Surfel-Based SLAM using 3D Laser Range Data in Urban Environments

Tangent Convolutions for Dense Prediction in 3D

A LiDAR Point Cloud Generator: from a Virtual World to Autonomous Driving

SqueezeNext: Hardware-Aware Neural Network Design

Depth-aware CNN for RGB-D Segmentation

SO-Net: Self-Organizing Network for Point Cloud Analysis

SPLATNet: Sparse Lattice Networks for Point Cloud Processing

Dynamic Graph CNN for Learning on Point Clouds

PointCNN: Convolution On X-Transformed Points

MobileNetV2: Inverted Residuals and Linear Bottlenecks

Large-Scale Point Cloud Semantic Segmentation with Superpoint Graphs

Frustum PointNets for 3D Object Detection from RGB-D Data

Shift: A Zero FLOP, Zero Parameter Alternative to Spatial Convolutions

VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection

SqueezeSeg: Convolutional Neural Nets with Recurrent CRF for Real-Time Road-Object Segmentation from 3D LiDAR Point Cloud

SEGCloud: Semantic Segmentation of 3D Point Clouds

Squeeze-and-Excitation Networks

3DCNN-DQN-RNN: A Deep Reinforcement Learning Framework for Semantic Parsing of Large-Scale 3D Point Clouds

ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Escape from Cells: Deep Kd-Networks for the Recognition of 3D Point Cloud Models

Multi-stage Multi-recursive-input Fully Convolutional Networks for Neuronal Boundary Detection

Deformable Convolutional Networks

SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

Multi-view 3D Object Detection Network for Autonomous Driving

OctNet: Learning Deep 3D Representations at High Resolutions

Perceptual Losses for Real-Time Style Transfer and Super-Resolution

Stacked Hourglass Networks for Human Pose Estimation

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1MB model size

Dark Memory and Accelerator-Rich System Optimization in the Dark Silicon Era

Deep Sliding Shapes for Amodal 3D Object Detection in RGB-D Images

Microsoft COCO: Common Objects in Context

Vision meets robotics: The KITTI dataset

Learning Multiple Layers of Features from Tiny Images

Field of Study

Venue Information

Name

Type

URL

Alternate Names