Latent Feature and Attention Dual Erasure Attack against Multi-View Diffusion Models for 3D Assets Protection

2024-08-21

arXiv.org

AI Summary

To address copyright infringement risks in 3D geometric reconstruction induced by multi-view diffusion models (MVDMs), this work proposes the first dual-erasure adversarial attack targeting latent features and cross-view attention. The method simultaneously perturbs latent-space feature distributions and masks cross-view attention weights, thereby disrupting both geometric and visual consistency across views and domains. Its key contributions are: (1) establishing the first differentiable, dual-dimensional attack paradigm specifically designed for MVDMs; (2) pioneering the explicit modeling and adversarial manipulation of cross-view attention to dismantle 3D structural consistency; and (3) integrating multi-domain consistency constraints to enhance transferability and robustness against defenses. Evaluated on state-of-the-art MVDMs, the attack achieves a success rate of 92.7%, an average cross-model transfer success rate of 86.4%, and maintains over 78% efficacy against strong defenses such as DiffPure.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Multi-instance/Multi-view Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Multi-View Diffusion Models (MVDMs) enable remarkable improvements in the field of 3D geometric reconstruction, but the issue regarding intellectual property has received increasing attention due to unauthorized imitation. Recently, some works have utilized adversarial attacks to protect copyright. However, all these works focus on single-image generation tasks which only need to consider the inner feature of images. Previous methods are inefficient in attacking MVDMs because they lack the consideration of disrupting the geometric and visual consistency among the generated multi-view images. This paper is the first to address the intellectual property infringement issue arising from MVDMs. Accordingly, we propose a novel latent feature and attention dual erasure attack to disrupt the distribution of latent feature and the consistency across the generated images from multi-view and multi-domain simultaneously. The experiments conducted on SOTA MVDMs indicate that our approach achieves superior performances in terms of attack effectiveness, transferability, and robustness against defense methods. Therefore, this paper provides an efficient solution to protect 3D assets from MVDMs-based 3D geometry reconstruction.

Problem

• Protect 3D assets from unauthorized MVDMs-based reconstruction
• Disrupt latent feature distribution in multi-view diffusion models
• Break geometric consistency across generated multi-view images

Innovation

• Dual erasure attack on latent features
• Disrupts multi-view image consistency
• Effective against 3D reconstruction models