3DMAD (3D Mask Attack Dataset)

3D 面具攻击数据库 (3DMAD) 是一个生物特征（面部）欺骗数据库。它包含 17 个人的 76500 帧，使用 Kinect 记录，用于真实访问和欺骗攻击。每个框架包括：
深度图像（640x480 像素 - 1x11 位）
对应的 RGB 图像（640x480 像素 - 3x8 位）
手动注释的眼睛位置（相对于 RGB 图像）。
在 3 个不同的会话中为所有受试者收集数据，每个会话捕获 5 个 300 帧的视频。录音是在受控条件下进行的，具有正面视图和中性表情。前两个会话专门用于真实访问样本，其中记录受试者的时间延迟约为 2 周。在第三个会话中，3D 掩码攻击由单个操作员（攻击者）捕获。

The 3D Mask Attack Database (3DMAD) is a biometric (facial) spoofing database. It contains 76,500 frames from 17 individuals, recorded using a Kinect for both genuine access and spoofing attacks. Each frame includes:
- Depth image (640×480 pixels, 1×11 bits)
- Corresponding RGB image (640×480 pixels, 3×8 bits)
- Manually annotated eye positions (relative to the RGB image).

Data was collected for all subjects across 3 distinct sessions, with each session capturing 5 videos of 300 frames each. Recordings were conducted under controlled conditions, with frontal views and neutral facial expressions. The first two sessions were exclusively dedicated to genuine access samples, with an approximate 2-week time gap between the two genuine sessions. In the third session, 3D mask spoofing attacks were captured by a single operator (attacker).