【数据集介绍】3D Mask Attack Dataset（3DMAD数据集）

【编辑时间】2018.09.16

【数据集介绍】3D Mask Attack Dataset（3DMAD数据集）

【数据集链接】http://www.idiap.ch/dataset/3dmad

 

概述

该数据库由76500帧17人组成，共255个视频(每个视频300帧)，使用Kinect为真正的子目标和掩膜攻击记录。每个帧包括一幅深度图像、相应的RGB图像和手动标注的眼睛位置。

 

1. Introduction（介绍）

This directory contains the data for the 3D Mask Attack Database. It consists of 76500 frames of 17 persons, recorded using Kinect.

【此目录包含3D掩码攻击数据库的数据。它由17个人组成的76500帧，使用Kinect记录。】

 

2. Database

============

The database is collected in 3 different sessions for all subjects and for each session 5 videos of 300 frames are captured. The recordings are done under controlled conditions, with frontal-view and neutral expression.

【该数据库被收集在3个不同的sessions中为所有的人，而且每个session有5个300帧的视频3被捕获。记录是在受控制的条件下进行的，包括正面观察和 中性表达】

The first two sessions are dedicated to the real access samples, in which subjects are recorded with a time delay of ~2 weeks between the acquisitions.In the third session, 3D mask attacks are captured.

 

【前两次会议专门讨论实际的访问样本，在这些样本中，记录对象的时间延迟为2周。

在第三节中，捕获3D掩码攻击。】

For each subject, following are available:

* 3 sessions x 5 videos x 300 frames

* Each frame consists of : 640x480 8-bit RGB image and 640x480 11-bit depth image

* Eye positions for each frame with respect to the RGB image

【关于每一主题，可获得以下资料：

* 3 sessions x 5 videos x 300 frames

* 每个帧由640x480 8位rgb图像和640x480 11位深度图像组成。

*每个帧相对于rgb图像的眼睛位置

】

3. Files

==============

 

session1.tar.gz - Videos and annotations from the first session

 

session1.tar.gz - Videos and annotations from the second session

 

session1.tar.gz - Videos and annotations from the third session

 

documentation.tar.gz - README file and a python script (raw_to_AVI.py) to demonstrate how to access the data and visualize it.

【自述文件和python脚本(RAW_to_AVI.py)，演示如何访问数据并将其可视化】

4. Protocol（协议）

==============

 

The subjects in the database are divided into 3 randomly chosen non-overlapping sets for training, development and testing. Number of identities assigned for each set are 7, 5 and 5, respectively. While experimenting on the database, it is recommended that training (e.g. for parameter optimization, building universal models) is done with training and development sets, if needed. The test set should be SOLELY used to report performances.

【将数据库中的subject随机分为3组，进行训练、开发和测试。为每组分配的数目分别为7、5和5。（共有17个人）】

 

An additional protocol is defined to measure the vulnerability to spoofing for which the development and test sets are further divided into enrollment and probe partitions. 

【此外，还定义了一种协议，开发和测试集进一步分为注册分区和探测分区，用于度量的欺骗漏洞。】While the first sessions are assigned as enrolled gallery samples, the second and third sessions are used for real and mask probing, respectively.

【当第一会话被指定为已注册的图片库样本时，第二和第三会话分别用于真实和掩码探测。】

 

5、作者信息  

Nesli Erdogmus

09/July/2013