Building Large Scale 3D Face Database for Face Analysis

Abstract

We propose to build a large scale 3D face database with dense correspondence for variant face analysis research purposes. Large scale means that the number of subjects in the database is more than 400, which is, to our best knowledge, the biggest1 one at this time. 3D face means that we provide both the texture and shape of human faces, which is also balanced in gender and race. Dense correspondence means that the key facials points with semantic meanings are carefully labeled and aligned among different faces, which can be used for a broad range of face analysis tasks. We provide the data description, data collection schema and the post-processing methods to help the usage of the data and future extension. More and more data is still being collected and processed to enlarge the extensive 3D face database. The proposed face database provides solid ground truth for human face related tasks such as alignment, tracking, recognition and animation, etc.