Data pertaining to Chapter 3 of the PhD dissertation: "Defending Gaze Estimation against Backdoor Attacks"

This dataset accompanies the paper <em>“</em>SecureGaze: Defending Gaze Estimation Against Backdoor Attacks<em>” </em>(Chapter 3 of the PhD dissertation). This research focuses on defending deep learning-based gaze estimation models from backdoor attacks. It aims to reverse-engineer the trigger function with a generative model for reliable backdoor detection and fine-tune the identified backdoored model with the reverse-engineered trigger function. This dataset contains the codes to reproduce the results presented in the paper and the real-time video demo.

本数据集随附于学术论文《SecureGaze：抵御后门攻击的视线估计方法》（该论文为博士学位论文第三章）。本研究聚焦于抵御基于深度学习的视线估计模型所遭受的后门攻击，其核心目标为借助生成模型逆向推导触发函数，以实现可靠的后门检测，并利用该逆向推导得到的触发函数对已识别的被植入后门的模型进行微调。本数据集包含复现论文中所述实验结果所需的代码，以及实时视频演示程序。