Acted Facial Expressions In The Wild

Quality data recorded in varied realistic environments is vital for effective human face related research. Currently available datasets for human facial expression analysis have been generated in highly controlled lab environments. We present a new dynamic 2D facial expressions database based on movies capturing diverse scenarios. A new XML schema based approach has been developed for the database collection and distribution tools. Realistic face data plays a vital role in the research advancement of facial expression analysis systems.
We have named our database Acted Facial Expressions in the Wild similar to the spirit of the Labeled Faces in the Wild (LFW) database. It contains 957 videos in AVI format labelled with six basic expressions Angry, Happy, Disgust, Fear, Sad, Surprise and the Neutral expression. We also wanted to capture the information on how facial expressions evolved in subjects with age. Therefore we have chosen sets of movies featuring the same actors. For example, the Harry Potter series forms a good platform to analyse how facial expressions of subjects evolve with age. We used thirty-seven movies from a diverse range of movie genres so as to cover as much varied expressions and natural environments as possible.
Much progress has been made in the fields of face recognition and human activity recognition in the past years due to the availability of realistic databases as well as robust representation and classification techniques. Inspired by them, we present a labelled temporal facial expression database from movies. Human facial expression databases till now have been captured in controlled ‘lab’ environments.

在多样化真实场景中采集的高质量人脸数据，对于高效开展人类面部相关研究至关重要。当前已公开的人类面部表情分析数据集，均生成于高度受控的实验室环境中。我们构建了一个全新的动态二维面部表情数据库，其数据源自涵盖多样场景的电影片段。针对该数据库的采集与分发工具，我们开发了一种基于可扩展标记语言（XML，eXtensible Markup Language）的新型架构方案。真实人脸数据对于面部表情分析系统的研究进展具有核心支撑作用。

我们将该数据库命名为野外演绎面部表情数据库（Acted Facial Expressions in the Wild，AFEW），其设计理念与野外标注人脸数据库（Labeled Faces in the Wild，LFW）一脉相承。该数据库包含957段AVI格式视频，标注涵盖六种基本表情：愤怒（Angry）、高兴（Happy）、厌恶（Disgust）、恐惧（Fear）、悲伤（Sad）、惊讶（Surprise）以及中性表情。我们同时希望捕捉受试者面部表情随年龄演变的相关信息，因此选取了多部包含同一演员的电影片段作为数据源。例如，《哈利·波特》系列电影便是分析受试者面部表情随年龄变化的优质研究平台。我们共选用了涵盖多种电影类型的37部影片，以期尽可能覆盖多样的表情表现与自然的拍摄环境。

近年来，得益于真实可用的数据库资源以及鲁棒的表征与分类技术，人脸识别与人类行为识别领域已取得诸多突破性进展。受此启发，我们构建了一个源自电影的标注型时序面部表情数据库。截至目前，已公开的人类面部表情数据库均采集于受控的‘实验室’环境中。

搜集汇总

背景与挑战

背景概述

该数据集是一个动态2D面部表情数据库，包含957个标注视频，覆盖六种基本表情和中性表情，数据来源于多样化的电影场景，旨在支持面部表情分析研究。