Computational Privacy for Computer Vision

Photons JPG at Pexels

Computer Vision applications are making a remarkable impact on society and advancing progress in Machine Learning. Our project aims to design Computer Vision algorithms that can learn and infer from private data. This will significantly empower industries where data can be sensitive - digital health, smart cameras, etc.

This year at ECCV'22, we present two research works in this direction -

1. Sanitizer: Protecting sensitive information in task-agnostic data release
In Sanitizer, we develop techniques that enable sharing of sensitive images for health imaging, forensics, etc. without compromising individual privacy

2. CBNS: Censoring By Noisy Sampling
In CBNS, we propose an end-to-end privacy-aware system for sharing 3D point cloud data that underlies several computer vision tasks such as 3D mapping, localization, etc.