WMCA 人脸检测数据集

WMCA 全称 Wide Multi Channel Presentation Attack，是一个检测人脸识别中攻击表现的数据集。人脸识别易受攻击，如使用硅胶面具等攻击阻碍人脸识别。随着这些攻击工具升级，仅靠视觉光谱检测无法应对挑战。因此产生了一种基于多通道卷积神经网络的人脸识别攻防数据集。该数据集由 1941 个来自 72 个不同人的真实简短视频组成。数据从几个通道进行拍摄记录（颜色、深度、红外线和热量）。数据集是在 Idiap “IARPA BATL” 和 “H2020 TESLA” 项目框架内制作的，目的是研究人脸识别攻防方法。

WMCA, whose full name is Wide Multi Channel Presentation Attack, is a dataset dedicated to detecting presentation attacks against face recognition systems. Face recognition systems are susceptible to diverse presentation attacks, such as those utilizing silicone masks that can bypass or disrupt normal face recognition workflows. As attack technologies continue to evolve, relying solely on visual-spectrum-based detection approaches can no longer effectively mitigate these escalating threats. To address this critical challenge, the WMCA face recognition presentation attack and defense dataset, tailored for multi-channel convolutional neural network (CNN)-based research, was developed. The dataset comprises 1941 authentic short video clips collected from 72 unique individuals. Data was captured across four modalities: color, depth, infrared, and thermal imaging. It was constructed within the frameworks of the Idiap "IARPA BATL" and "H2020 TESLA" projects, with the core objective of advancing research on face recognition presentation attack and defense methodologies.

搜集汇总

背景与挑战

背景概述

WMCA人脸检测数据集是一个多通道的人脸识别攻防数据集，包含1941个来自72人的真实简短视频，涵盖颜色、深度、红外线和热量等多种拍摄通道。该数据集旨在研究人脸识别中的攻击检测方法，由Idiap研究所在IARPA BATL和H2020 TESLA项目框架内制作。