Robustifying Point Cloud Networks by Refocusing (2023-08-10T00:00:00.000000Z)

TL;DR

This study develops a general mechanism to increase point clouds neural networks robustness based on focus analysis and proposes a parameter-free refocusing algorithm that aims to unify all corruptions under the same distribution.

Abstract

The ability to cope with out-of-distribution (OOD) corruptions and adversarial attacks is crucial in real-world safety-demanding applications. In this study, we develop a general mechanism to increase point clouds neural networks robustness based on focus analysis. Recent studies have revealed the phenomenon of Overfocusing, which leads to a performance drop. When the network is primarily influenced by small input regions, it becomes less robust and prone to misclassify under noise and corruptions. However, quantifying overfocusing is still vague and lacks clear definitions. Here, we provide a mathematical definition of focus, overfocusing and underfocusing. The notions are general, but in this study, we specifically investigate the case of 3D point clouds. We observe that corrupted sets result in a biased focus distribution compared to the clean training set. We show that as focus distribution deviates from the one learned in the training phase - classification performance deteriorates. We thus propose a parameter-free refocusing algorithm that aims to unify all corruptions under the same distribution. We validate our findings on a 3D zero-shot classification task, achieving SOTA in robust 3D classification on ModelNet-C dataset, and in adversarial defense against Shape-Invariant attack.

Authors

Meir Yossef Levi

1 papers

Guy Gilboa

1 papers

TL;DR

Abstract

Authors

References67 items

POINTACL: Adversarial Contrastive Learning for Robust Point Clouds Representation Under Adversarial Attack

Self-Ordering Point Clouds

EPiC: Ensemble of Partial Point Clouds for Robust Classification

Robustifying Token Attention for Vision Transformers

PointCert: Point Cloud Classification with Deterministic Certified Robustness Guarantees

An Improved Explainable Point Cloud Classifier (XPCC)

Explanation on Pretraining Bias of Finetuned Vision Transformer

PointACL: Adversarial Contrastive Learning for Robust Point Clouds Representation under Adversarial Attack

Shape-invariant 3D Adversarial Point Clouds

LPF-Defense: 3D adversarial defense based on frequency analysis

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

Benchmarking and Analyzing Point Cloud Classification under Corruptions

PAGN: perturbation adaption generation network for point cloud adversarial defense

CloudWalker: Random walks for 3D point cloud shape analysis

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

Explainability-Aware One Point Attack for Point Cloud Neural Networks

Point Cloud Augmentation with Weighted Local Transformations

Surrogate Model-Based Explainability Methods for Point Cloud NNs

Less is More: Pay Less Attention in Vision Transformers

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

PAConv: Position Adaptive Convolution with Dynamic Kernel Assembling on Point Clouds

PointGuard: Provably Robust 3D Point Cloud Classification

Regularization Strategy for Point Cloud via Rigidly Mixed Sample

PointCutMix: Regularization Strategy for Point Cloud Classification

Learning Geometry-Disentangled Representation for Complementary Understanding of 3D Object Point Cloud

PCT: Point cloud transformer

Transformer Interpretability Beyond Attention Visualization

IF-Defense: 3D Adversarial Point Cloud Defense via Implicit Function based Restoration

Minimal Adversarial Examples for Deep Learning on 3D Point Clouds

LG-GAN: Label Guided Adversarial Network for Flexible Targeted Attack of Point Cloud Based Deep Networks

ShapeAdv: Generating Shape-Aware Adversarial 3D Point Clouds

Shortcut learning in deep neural networks

Robust Adversarial Objects against Deep Learning Models

3D Point Cloud Feature Explanations Using Gradient-Based Methods

PointASNL: Robust Point Clouds Processing Using Nonlocal Neural Networks With Adaptive Sampling

On Isometry Robustness of Deep 3D Point Cloud Models Under Adversarial Attacks

SampleNet: Differentiable Point Cloud Sampling

AdvPC: Transferable Adversarial Perturbations on 3D Point Clouds

On the Relationship between Self-Attention and Convolutional Layers

Adversarial Shape Perturbations on 3D Point Clouds

PointHop: An Explainable Machine Learning Method for Point Cloud Classification

Natural Adversarial Examples

Analyzing the Structure of Attention in a Transformer Language Model

Robustness of 3D Deep Learning in an Adversarial Setting

Adversarial Attack and Defense on Point Sets

Extending Adversarial Attacks and Defenses to Deep 3D Point Cloud Classifiers

PointCleanNet: Learning to Denoise and Remove Outliers from Dense Point Clouds

DUP-Net: Denoiser and Upsampler Network for 3D Adversarial Point Clouds Defense

PointCloud Saliency Maps

Generating 3D Adversarial Point Clouds

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

Dynamic Graph CNN for Learning on Point Clouds

What does Attention in Neural Machine Translation Pay Attention to?

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

Ensemble Adversarial Training: Attacks and Defenses

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

Multi-view 3D Object Detection Network for Autonomous Driving

Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization

Shape completion enabled robotic grasping

On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation

3D ShapeNets: A deep representation for volumetric shapes

Towards 3D Point cloud based object maps for household environments

BubblEX: An Explainable Deep Learning Framework for Point-Cloud Classification

Learning to Sample

INDICES OF QUALITATIVE VARIATION.

agreement with IEEE. Restrictions apply

We provide a general definition for the network’s focus and for over-and under-focusing. A comprehensive analysis is performed relating corruptions to the focus distribution

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names