Summary of Hiding-in-plain-sight (hips) Attack on Clip For Targetted Object Removal From Images, by Arka Daw et al.
Hiding-in-Plain-Sight (HiPS) Attack on CLIP for Targetted Object Removal from Images
by Arka Daw, Megan Hong-Thanh Chung, Maria Mahbub, Amir Sadovnik
First submitted to arxiv on: 16 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 paper proposes a novel class of adversarial attacks called Hiding-in-Plain-Sight (HiPS) attacks that can subtly modify the predictions of large multi-modal models like CLIP. These attacks aim to conceal target objects in images and remove them from captions, making the changes undetectable by downstream models or humans. The authors introduce two HiPS attack variants: HiPS-cls and HiPS-cap, which are demonstrated to be effective in transferring to downstream image captioning models like CLIP-Cap. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new way for bad guys to trick big computer vision models into hiding things from images and captions. They call this “Hiding-in-Plain-Sight” attacks. These attacks can make it hard for other computers or even humans to notice when something is missing from an image or caption. The researchers tested two types of these attacks on a model called CLIP-Cap, which writes descriptions of images. |
Keywords
» Artificial intelligence » Image captioning » Multi modal
