Summary of Adversarial Attacks Using Differentiable Rendering: a Survey, by Matthew Hull et al.
Adversarial Attacks Using Differentiable Rendering: A Survey
by Matthew Hull, Chao Zhang, Zsolt Kira, Duen Horng Chau
First submitted to arxiv on: 14 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The abstract discusses recent advancements in differentiable rendering methods, which can generate photo-realistic and physically plausible adversarial attacks on deep neural networks (DNNs). Specifically, it highlights the emergence of libraries like Mitsuba, PyTorch3D, and techniques like Neural Radiance Fields and 3D Gaussian Splatting for solving inverse rendering problems. The abstract notes that the research community has not yet fully explored these capabilities for generating attacks, due to varying attack goals (misclassification, misdetection) and tasks (manipulating textures, altering illumination, modifying 3D meshes). A task-oriented unifying framework is proposed to summarize existing works, reveal research gaps, and identify future directions. The survey focuses on how various tasks enable attacks on different DNNs (image classification, facial recognition, object detection, optical flow, depth estimation), highlighting vulnerabilities in computer vision systems against photorealistic adversarial attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about a new way to make fake images that can trick artificial intelligence. It talks about special techniques called differentiable rendering methods that can create super-realistic pictures and videos by manipulating 3D objects and scenes. These fake images are called “adversarial attacks” because they can fool AI systems into making wrong decisions. The paper says that many researchers have been working on this topic, but there is still a lot to learn about how these techniques can be used to attack different types of AI systems. |
Keywords
» Artificial intelligence » Depth estimation » Image classification » Object detection » Optical flow
