PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (2022-06-09T00:00:00.000000Z)

TL;DR

This work revisits the classical PointNet++ through a systematic study of model training and scaling strategies, and proposes PointNeXt, the next version of PointNets, which can be flexibly scaled up and outperforms state-of-the-art methods on both 3D classification and segmentation tasks.

Abstract

PointNet++ is one of the most influential neural architectures for point cloud understanding. Although the accuracy of PointNet++ has been largely surpassed by recent networks such as PointMLP and Point Transformer, we find that a large portion of the performance gain is due to improved training strategies, i.e. data augmentation and optimization techniques, and increased model sizes rather than architectural innovations. Thus, the full potential of PointNet++ has yet to be explored. In this work, we revisit the classical PointNet++ through a systematic study of model training and scaling strategies, and offer two major contributions. First, we propose a set of improved training strategies that significantly improve PointNet++ performance. For example, we show that, without any change in architecture, the overall accuracy (OA) of PointNet++ on ScanObjectNN object classification can be raised from 77.9% to 86.1%, even outperforming state-of-the-art PointMLP. Second, we introduce an inverted residual bottleneck design and separable MLPs into PointNet++ to enable efficient and effective model scaling and propose PointNeXt, the next version of PointNets. PointNeXt can be flexibly scaled up and outperforms state-of-the-art methods on both 3D classification and segmentation tasks. For classification, PointNeXt reaches an overall accuracy of 87.7 on ScanObjectNN, surpassing PointMLP by 2.3%, while being 10x faster in inference. For semantic segmentation, PointNeXt establishes a new state-of-the-art performance with 74.9% mean IoU on S3DIS (6-fold cross-validation), being superior to the recent Point Transformer. The code and models are available at https://github.com/guochengqian/pointnext.

Authors

References61 items

Surface Representation for Point Clouds

PolyLoss: A Polynomial Expansion Perspective of Classification Loss Functions

Stratified Transformer for 3D Point Cloud Segmentation

Contrastive Boundary Learning for Point Cloud Segmentation

Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework

A ConvNet for the 2020s

Fast Point Transformer

Point-BERT: Pre-training 3D Point Cloud Transformers with Masked Point Modeling

PatchFormer: An Efficient Point Transformer with Patch Attention

ASSANet: An Anisotropic Separable Set Abstraction for Efficient Point Cloud Representation Learning

ResNet strikes back: An improved training procedure in timm

Training Graph Neural Networks with 1000 Layers

Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline

Scaling Vision Transformers

Walk in the Cloud: Learning Curves for Point Clouds Shape Analysis

PRA-Net: Point Relation-Aware Network for 3D Point Cloud Analysis

Revisiting ResNets: Improved Training and Scaling Strategies

Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusion

PCT: Point cloud transformer

MVTN: Multi-View Transformation Network for 3D Shape Recognition

Point Transformer

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

A Closer Look at Local Aggregation Operators in Point Cloud Analysis

Dense-Resolution Network for Point Cloud Classification and Segmentation

PU-GCN: Point Cloud Upsampling using Graph Convolutional Networks

Geometric Back-Projection Network for Point Cloud Classification

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

DeepGCNs: Making GCNs Go as Deep as CNNs

Revisiting Point Cloud Classification: A New Benchmark Dataset and Classification Model on Real-World Data

Point-Voxel CNN for Efficient 3D Deep Learning

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

KPConv: Flexible and Deformable Convolution for Point Clouds

4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks

Relation-Shape Convolutional Neural Network for Point Cloud Analysis

DeepGCNs: Can GCNs Go As Deep As CNNs?

PointConv: Deep Convolutional Networks on 3D Point Clouds

Tangent Convolutions for Dense Prediction in 3D

SpiderCNN: Deep Learning on Point Sets with Parameterized Convolutional Filters

Dynamic Graph CNN for Learning on Point Clouds

PointCNN: Convolution On X-Transformed Points

MobileNetV2: Inverted Residuals and Linear Bottlenecks

3D Semantic Segmentation with Submanifold Sparse Convolutional Networks

Decoupled Weight Decay Regularization

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

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

ScanNet: Richly-Annotated 3D Reconstructions of Indoor Scenes

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

A scalable active framework for region annotation in 3D shape collections

3D Semantic Parsing of Large-Scale Indoor Spaces

Deep Residual Learning for Image Recognition

ShapeNet: An Information-Rich 3D Model Repository

Rethinking the Inception Architecture for Computer Vision

U-Net: Convolutional Networks for Biomedical Image Segmentation

Multi-view Convolutional Neural Networks for 3D Shape Recognition

Adam: A Method for Stochastic Optimization

Going deeper with convolutions

3D ShapeNets: A deep representation for volumetric shapes

ImageNet: A large-scale hierarchical image database

Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs

Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

Pointnet/pointnet++ pytorch. https://github.com/yanx27/Pointnet_Pointnet2_ pytorch, 2019