Date: November 2023.
Source: Pattern Recognition Letters, Volume 175, Pages 23-29, ISSN 0167-8655, https://doi.org/10.1016/j.patrec.2023.09.015.
Research Summary: Existing 3D face datasets are typically built using either high-resolution scanners or lower-cost consumer devices like the Kinect, but few combine both to study the gap between them. Florence Multi-Resolution 3D Facial Expression (Florence 3DMRE) addresses this by pairing synchronized high-resolution (HR) and Kinect-based low-resolution (LR) scans of the same subjects and expressions. The dataset includes 14 real-world subjects, each performing 19 complex, asymmetric facial expressions modeled on facial rehabilitation exercises, with each expression captured as both an HR scan and an RGB-D sequence. The authors use the dataset to establish baseline results for cross-resolution 3D face recognition and reconstruction, highlighting it as an open research challenge.
3dMD’s Role: Florence 3DMRE’s high-resolution (HR), real-world 3D facial scans were captured on a static 3dMDface System. This is the third confirmed 3dMD-based dataset from the same lab, following the original 2011 Florence 2D/3D dataset and the 2022 Florence 4D Facial Expression Dataset.
Article: The Florence multi-resolution 3D facial expression dataset.
Authors: Claudio Ferrari, Stefano Berretti, Pietro Pala, Alberto Del Bimbo. Department of Information Engineering, University of Florence, Florence, Italy