Go to ICML 2025 Conference homepageEgoPrivacy: What Your First-Person Camera Says About You?
TL;DR: We present EgoPrivacy, a comprehensive large-scale benchmark for evaluating the privacy issues in egocentric videos. We further propose Retrieval Augmented Attack, a novel attack that leaks the camera wearer's privacy information.
Abstract: While the rapid proliferation of wearable cameras has raised significant concerns about egocentric video privacy, prior work has largely overlooked the unique privacy threats posed to the camera wearer. This work investigates the core question: How much privacy information about the camera wearer can be inferred from their first-person view videos? We introduce EgoPrivacy, the first large-scale benchmark for the comprehensive evaluation of privacy risks in egocentric vision. EgoPrivacy covers three types of privacy (demographic, individual, and situational), defining seven tasks that aim to recover private information ranging from fine-grained (e.g., wearer's identity) to coarse-grained (e.g., age group). To further emphasize the privacy threats inherent to egocentric vision, we propose Retrieval-Augmented Attack, a novel attack strategy that leverages ego-to-exo retrieval from an external pool of exocentric videos to boost the effectiveness of demographic privacy attacks. An extensive comparison of the different attacks possible under all threat models is presented, showing that private information of the wearer is highly susceptible to leakage. For instance, our findings indicate that foundation models can effectively compromise wearer privacy even in zero-shot settings by recovering attributes such as identity, scene, gender, and race with 70–80% accuracy. Our code and data are available at https://github.com/williamium3000/ego-privacy.
Lay Summary: Recent wearable cameras—like those used in action cams, smart glasses, and body cams—are great at capturing what’s happening around us. But there’s another side we tend to forget: they also reveal a lot about you, the person wearing them. While most people worry about how others are seen and tracked in these videos, this paper shows that it’s actually just as risky for the person wearing the camera.
We introduce EgoPrivacy, the first large-scale set of wearable-camera videos that helps us measure just how much personal information can be taken from these videos. It’s more than just faces—it includes figuring out your age, your identity, and even details about your surroundings.
We also show a clever new method—called Retrieval-Augmented Attack—where a computer secretly looks through regular videos (like those from phones and webcams) to find matches and piece together who you are, just from what your camera sees.
Our tests reveal a surprising truth: wearable cameras are leaking a lot of private information about their wearers—even when you think the footage is harmless. That means as these devices become more common, we all need better ways to protect you, not just the people around you.
Link To Code: https://github.com/williamium3000/ego-privacy
Primary Area: Social Aspects->Privacy
Keywords: egocentric vision, privacy, retrieval
Submission Number: 9079
Loading