RenFace is the only large-scale whole human-head dataset with rich fine-grained hairstyles, as well as phoneme-balanced speech videos produced by the corresponding individual/head. The total number of IDs in this dataset is more than 500. The total number of captured frames reaches 80M, covering 60 viewing angles, with an image resolution of 2k. The dataset contains multi-ethnic, multi-age, multi-ethnic subjects, matched with multiple high fine-grained hairstyles and hair colours, with audio and video collected simultaneously on a reading corpus. The dynamic 4D face dataset can be applied in the fields of 2D/3D digital human face generation, face reconstruction and face capture etc..
Key Features:
Large scale data (IDs>500, Frames>80M)
Various facial expressions (natural expressions plus 11 exaggerated expressions)
Various hair colours (natural hairstyles/hair colours plus special hair colours and head covers)
Phoneme-balanced corpus
Various viewings of the head captures (60)
HD images (2k)
Rich annotations (matting, camera params, landmarks, flame params, appearance label, text description, action units)
Annotation types: Camera Calibration, Mask, Head Parsing 2D & 3D landmarks, Scan
Application scene: Human Face Reconstruction
Sensitive data type: Human Face
