CelebFaces Attributes (CelebA) Dataset

CelebFaces Attributes (CelebA) Dataset

原文：

CelebFaces Attributes (CelebA) Dataset

Over 200k images of celebrities with 40 binary attribute annotations

A popular component of computer vision and deep learning revolves around identifying faces for various applications from logging into your phone with your face or searching through surveillance images for a particular suspect. This dataset is great for training and testing models for face detection, particularly for recognising facial attributes such as finding people with brown hair, are smiling, or wearing glasses. Images cover large pose variations, background clutter, diverse people, supported by a large quantity of images and rich annotations. This data was originally collected by researchers at MMLAB, The Chinese University of Hong Kong (specific reference in Acknowledgment section).

Overall

202,599 number of face images of various celebrities10,177 unique identities, but names of identities are not given40 binary attribute annotations per image5 landmark locations

Data Files

imgalignceleba.zip: All the face images, cropped and alignedlistevalpartition.csv: Recommended partitioning of images into training, validation, testing sets. Images 1-162770 are training, 162771-182637 are validation, 182638-202599 are testinglistbboxceleba.csv: Bounding box information for each image. "x1" and "y1" represent the upper left point coordinate of bounding box. "width" and "height" represent the width and height of bounding boxlistlandmarksalign_celeba.csv: Image landmarks and their respective coordinates. There are 5 landmarks: left eye, right eye, nose, left mouth, right mouthlistattrceleba.csv: Attribute labels for each image. There are 40 attributes. "1" represents positive while "-1" represents negative

译文：

CelebFaces属性（CelebA）数据集

超过20万张名人图片，带有40个二进制属性注释

计算机视觉和深度学习的一个流行组成部分围绕着识别各种应用程序的人脸展开，这些应用程序包括用你的脸登录你的手机，或者在监控图像中搜索特定嫌疑人。这个数据集对于训练和测试人脸检测模型非常有用，特别是对于识别人脸属性，例如发现棕色头发、微笑或戴眼镜的人。图像涵盖了姿势变化大、背景杂乱、人的多样性，并有大量的图像和丰富的注释支持。这一数据最初是由香港大学MMLRA的研究人员收集的（具体参考文献部分）。

总体

●202599各种名人的面部图像数量

●10177个唯一标识，但未给出标识名称

●每幅图像40个二进制属性注释

●5个地标位置

数据文件

● imgalignceleba.zip文件：所有面部图像，裁剪并对齐

● ListValPartition.csv文件：建议将图像划分为训练集、验证集和测试集。图片1-162770为培训，162771-182637为验证，182638-202599为测试

● listbboxceleba.csv文件：每个图像的边界框信息。“x1”和“y1”表示边界框的左上角点坐标“宽度”和“高度”表示边界框的宽度和高度

●listlandmarksalign_celeba.csv：图像地标及其各自的坐标。有5个地标：左眼、右眼、鼻子、左嘴、右嘴

● listattrceleba.csv文件：每个图像的属性标签。共有40个属性。”1“表示正，-1”表示负

​

版权声明：本文内容由网络用户投稿，版权归原作者所有，本站不拥有其著作权，亦不承担相应法律责任。如果您发现本站中有涉嫌抄袭或描述失实的内容，请联系我们jiasou666@gmail.com 处理，核实后本网站将在24小时内删除侵权内容。