Two papers accepted at CVPR 2026.
PapersPrivacy-preserving vision
Shielding faces and personal images from unauthorized generative use.
Research
Safeguarding visual identity in the age of generative models.
Concept erasure
Removing unsafe concepts from generative models, and making it stick.
Reliable watermarking
Testing which invisible marks survive, and what quietly breaks them.
Model safety
Defenses against malicious editing and unauthorized personalization.
Open to collaborations in generative AI safety and trustworthy vision — get in touch.
Selected news
Recent work and recognition.
2026
FaceGuardian accepted at SaTML 2026.
OralSelected for the SaTML travel grant.
Grant2025
STEREO accepted at CVPR 2025.
HighlightOutstanding Reviewer, CVPR 2025.
AwardTrojanWave accepted at EMNLP 2025.
OralFace Anonymixer accepted at IJCB 2025.
SpotlightSelected for the CVPR 2025 travel grant.
Grant2024
Won the NeurIPS 2024 Invisible Watermark Removal Challenge — black-box and beige-box tracks.
1st placeBest Poster, MBZUAI Research Symposium.
AwardBaPLe and PromptSmooth accepted at MICCAI 2024.
PapersSelected for the DALL·E Detection Classifier Access Program.
Access