MaskedFace-Net

MaskedFace-Net 是基于数据集 Flickr-Faces-HQ (FFHQ) 的带有正确或错误佩戴面具的人脸数据集（133,783 张图像）。戴口罩似乎是限制 COVID-19 传播的一种解决方案。在这种情况下，需要有效的识别系统来检查人脸是否在受监管的区域被遮盖。要执行此任务，需要大量蒙面人脸数据集来训练深度学习模型以检测戴口罩和未戴口罩的人。文献中提供了一些大型蒙面人脸数据集。然而，目前，没有可用的大型蒙面人脸图像数据集可以检查检测到的蒙面人脸是否正确佩戴。事实上，由于不良做法、不良行为或个人（例如儿童、老人）的脆弱性，许多人没有正确佩戴口罩。出于这些原因，一些戴口罩运动旨在使人们对这个问题和良好做法敏感。从这个意义上说，这项工作提出了三种蒙面人脸检测数据集；即，正确蒙面人脸数据集（CMFD）、错误蒙面人脸数据集（IMFD）及其组合用于全局蒙面人脸检测（MaskedFace-Net）。提出了具有双重目标的真实蒙面人脸数据集：i) 检测人脸是否被蒙面，ii) 检测是否正确佩戴或错误佩戴口罩的人脸（例如，在机场入口处或人群中）。据我们所知，没有大型蒙面人脸数据集提供如此精细的分类以允许进行口罩佩戴分析。此外，这项工作在全球范围内展示了应用的蒙版到脸可变形模型，用于允许生成其他蒙版的人脸图像，特别是带有特定蒙版的图像。

MaskedFace-Net is a face dataset containing 133,783 images of faces with correctly or incorrectly worn masks, based on the Flickr-Faces-HQ (FFHQ) dataset. Wearing face masks is regarded as a solution to limit the spread of COVID-19. In this scenario, effective recognition systems are required to verify whether faces are covered in regulated areas. To accomplish this task, large-scale masked face datasets are needed to train deep learning models for detecting individuals who wear masks and those who do not. Although several large-scale masked face datasets have been presented in existing literature, currently there is no available large-scale masked face image dataset that can inspect whether detected masked faces are properly worn. In fact, due to improper practices, non-compliant behaviors, or the vulnerability of groups such as children and the elderly, many people fail to wear masks correctly. For these reasons, multiple mask-wearing campaigns have been launched to raise public awareness of this issue and proper mask-wearing protocols. In this regard, this work proposes three masked face detection datasets: the Correctly Masked Face Dataset (CMFD), the Incorrectly Masked Face Dataset (IMFD), and their combination for global masked face detection (MaskedFace-Net). The real-world masked face datasets proposed in this work have dual objectives: i) detecting whether a face is covered by a mask, and ii) detecting whether the mask is worn correctly or incorrectly (e.g., at airport entrances or in crowded scenarios). To the best of our knowledge, no large-scale masked face dataset provides such fine-grained classification to enable mask-wearing analysis. Additionally, this work demonstrates the global application of a face deformable model with applied masks, allowing the generation of additional masked face images, particularly those with specific types of masks.

搜集汇总

背景与挑战

背景概述

MaskedFace-Net是一个包含133,783张正确或错误佩戴口罩的人脸图像数据集，旨在支持COVID-19背景下的口罩佩戴检测研究。该数据集细分为正确佩戴口罩(CMFD)和错误佩戴口罩(IMFD)两类，可用于训练深度学习模型进行口罩佩戴及其正确性的检测。