Training and testing labelled dataset of human faces for different indoor visual comfort and glare visual discomfort situations - videos

The video dataset is generated from experimental tests on 23 different participants and includes a certain variety in characteristics: age, gender, face feature, glasses, etc. The original dataset was generated from experimental tests on 24 different participants but all the data of one of the participants was entirely excluded as this participant did not want his/her data to be publically available. After synchronization of the video footages with the answers of the participants regarding their visual discomfort at the different phases of the experiment to label the video dataset, the video footages are resampled at a 0.5 FPS rate, and the YOLOFace extraction algorithm is used to crop the image down to the face of the occupants only and generate the labelled image dataset.

本视频数据集源自针对23名不同受试者开展的实验测试，涵盖多维度特征差异：年龄、性别、面部特征、是否佩戴眼镜等。原始数据集最初采集自24名受试者的实验测试，但其中一名受试者因不愿公开其个人数据，其全部数据均被移除。为完成该视频数据集的标注工作，先将视频片段与受试者在实验各阶段反馈的视觉不适情况进行对齐，随后将视频帧率重采样至0.5 FPS，并使用YOLOFace提取算法仅裁剪出画面中的受试者面部区域，最终生成带标注的图像数据集。