Activity Recognition on Avatar-Anonymized Datasets with Masked Differential Privacy
This work addresses privacy concerns in video datasets for computer vision applications, offering a more efficient solution than standard methods, though it is incremental in combining existing techniques.
The paper tackled the problem of privacy-preserving computer vision by introducing an anonymization pipeline that replaces humans in videos with synthetic avatars and a masked differential privacy method to protect background information, resulting in better utility-privacy trade-offs, especially for ε<1, as shown on multiple action recognition datasets.
Privacy-preserving computer vision is an important emerging problem in machine learning and artificial intelligence. Prevalent methods tackling this problem use differential privacy (DP) or obfuscation techniques to protect the privacy of individuals. In both cases, the utility of the trained model is sacrificed heavily in this process. In this work, we present an anonymization pipeline that replaces sensitive human subjects in video datasets with synthetic avatars within context, employing a combined rendering and stable diffusion-based strategy. Additionally we propose masked differential privacy ({MaskDP}) to protect non-anonymized but privacy sensitive background information. MaskDP allows for controlling sensitive regions where differential privacy is applied, in contrast to applying DP on the entire input. This combined methodology provides strong privacy protection while minimizing the usual performance penalty of privacy preserving methods. Experiments on multiple challenging action recognition datasets demonstrate that our proposed techniques result in better utility-privacy trade-offs compared to standard differentially private training in the especially demanding $ε<1$ regime.