Reproduce “ADD: An Automatic Desensitization Fisheye Dataset for Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zizhang Wu, Chenxin Yuan, Hongyang Wei, Fan Song, Tianhao Xu
This work introduces ADD, the first large-scale dataset of 650,000 fisheye camera images specifically designed for desensitizing human faces and vehicle license plates in autonomous driving contexts. A multitask deep learning framework called DesCenterNet is also proposed, which integrates detection and segmentation to achieve real-time, precise privacy protection, outperforming existing methods.
Key insights
- The ADD dataset is the first large-scale collection of 650,000 fisheye camera images specifically annotated for face and license plate desensitization, addressing a critical gap for autonomous driving.
- The proposed DesCenterNet framework integrates a multitask detection module (CenterNet-based) and a segmentation module (FCN) to achieve precise, real-time desensitization.
- A specialized evaluation metric, Importance of Facial Features (IOFF), was introduced to assess desensitization quality by weighting different facial regions based on their contribution to recognition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.