Pre-training strategies and datasets for facial representation learning

30 Mar 2021  ยท  Adrian Bulat, Shiyang Cheng, Jing Yang, Andrew Garbett, Enrique Sanchez, Georgios Tzimiropoulos ยท

What is the best way to learn a universal face representation? Recent work on Deep Learning in the area of face analysis has focused on supervised learning for specific tasks of interest (e.g. face recognition, facial landmark localization etc.) but has overlooked the overarching question of how to find a facial representation that can be readily adapted to several facial analysis tasks and datasets. To this end, we make the following 4 contributions: (a) we introduce, for the first time, a comprehensive evaluation benchmark for facial representation learning consisting of 5 important face analysis tasks. (b) We systematically investigate two ways of large-scale representation learning applied to faces: supervised and unsupervised pre-training. Importantly, we focus our evaluations on the case of few-shot facial learning. (c) We investigate important properties of the training datasets including their size and quality (labelled, unlabelled or even uncurated). (d) To draw our conclusions, we conducted a very large number of experiments. Our main two findings are: (1) Unsupervised pre-training on completely in-the-wild, uncurated data provides consistent and, in some cases, significant accuracy improvements for all facial tasks considered. (2) Many existing facial video datasets seem to have a large amount of redundancy. We will release code, and pre-trained models to facilitate future research.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Valence Estimation AffectNet Ours (VGG-F) CCC 0.710 # 2
RMSE 0.356 # 2
Arousal Estimation AffectNet Ours (VGG-F) RMSE 0.326 # 2
CCC 0.629 # 2
Face Alignment AFLW-19 VGG-F NME_diag (%, Full) 1.55 # 10
3D Face Reconstruction AFLW2000-3D VGG-F Mean NME 3.42% # 2
Facial Expression Recognition (FER) BP4D Ours (VGG-F) ICC 0.719 # 1
Face Alignment COFW Ours (VGG-F) NME (inter-ocular) 3.32 # 9
Facial Expression Recognition (FER) DISFA Ours (VGG-F) ICC 0.598 # 1
Face Alignment WFW (Extra Data) VGG-F NME (inter-ocular) 4.57 # 11

Methods


No methods listed for this paper. Add relevant methods here