CelebAGaze

CelebAGaze 包含从 CelebA 和互联网收集的 25283 张高分辨率名人图像。它由21832张眼睛盯着相机的人脸图像和3451张眼睛盯着其他地方的人脸图像组成。裁剪所有图像（256 × 256）并计算 dlib 的眼罩区域。具体来说，dlib 用于提取 68 个面部标志并计算眼睛区域附近的 6 个点的平均值，这将是蒙版的中心点。蒙版的大小固定为 30×50。如上所述，随机选择来自 Y 域的 300 个样本，来自 X 域的 100 个样本作为测试集，其余作为训练集。请注意，此数据集是未配对的，并且未标有特定的眼睛角度或头部姿势信息。CelebAGaze contains 25,283 high-resolution celebrity images collected from CelebA and the Internet. It consists of 21,832 facial images where subjects gaze directly at the camera, and 3,451 facial images where subjects gaze elsewhere. All images are cropped to 256 × 256 resolution, and eye mask regions are calculated using dlib. Specifically, dlib is employed to extract 68 facial landmarks, and the average of 6 points near the eye regions is calculated as the center point of the mask. The size of the mask is fixed at 30×50. As previously stated, 300 samples from the Y domain and 100 samples from the X domain are randomly selected as the test set, with the remaining samples serving as the training set. Notably, this dataset is unpaired, and no specific eye angle or head pose information is annotated.

搜集汇总

背景与挑战

背景概述

CelebAGaze数据集包含25283张高分辨率名人图像，源自CelebA和互联网，其中21832张为注视相机的人脸，3451张为注视其他方向的人脸。所有图像均被裁剪为256×256，并利用dlib计算眼罩区域，划分为训练集和测试集，但未标注眼睛角度或头部姿势信息。