现实世界中的情感脸谱多标签数据集

人类的面部表现往往是以不同基本情绪的组合、混合或复合的形式出现。Real-world Affective Faces Multi Label (RAF-ML)是一个多标签的面部表情数据集，其中有大约5K张从互联网上下载的多样化的面部图像，这些图像具有混合的情绪，以及受试者的身份、头部姿势、照明条件和遮挡物的变化情况。 在注解过程中，315名训练有素的注解员被聘用，以确保每张图片都能被独立注解足够的次数。而具有多峰标签分布的图像被挑选出来，构成RAF-ML。 在RAF-ML中，我们提供了4908张具有混合情绪的真实世界图像，每张图像的6维表情分布向量，5个准确的地标位置和37个自动地标位置，以及多标签情绪识别的基线分类器输出。 For more details of the dataset, please refer to the paper "Blended Emotion in-the-wild: Multi-label Facial expression Recognition Using Crowdsourced Annotations and Deep Locality Feature Learning". If you use the RAF-ML datatset, please cite the paper below: Please contact Shan Li and Weihong Deng for questions about the database.

Human facial expressions often manifest as combinations, blends, or composites of distinct basic emotions. Real-world Affective Faces Multi Label (RAF-ML) is a multi-label facial expression dataset containing approximately 5,000 diverse facial images downloaded from the Internet. These images feature mixed emotions, alongside variations in subject identities, head poses, lighting conditions, and occlusions. During the annotation phase, 315 well-trained annotators were recruited to ensure that each image was independently annotated a sufficient number of times. Images with multi-modal label distributions were selected to form the RAF-ML dataset. For the RAF-ML dataset, we provide 4,908 real-world images with mixed emotions, accompanied by 6-dimensional emotion distribution vectors for each image, 5 accurate landmark positions, 37 automatically detected landmark positions, and baseline classifier outputs for multi-label emotion recognition. For more details about this dataset, please refer to the paper titled "Blended Emotion in-the-wild: Multi-label Facial Expression Recognition Using Crowdsourced Annotations and Deep Locality Feature Learning". If you use the RAF-ML dataset, please cite the aforementioned paper. For inquiries regarding the database, please contact Shan Li and Weihong Deng.

搜集汇总

背景与挑战

背景概述

现实世界中的情感脸谱多标签数据集（RAF-ML）是一个包含约5K张多样化面部图像的数据集，用于多标签面部表情识别。数据集提供了每张图像的6维表情分布向量、地标位置以及基线分类器输出，适用于非商业研究用途。