2020/04/14 by Ιωάννα Ντίνου, Ntinou, Ioanna, Enrique Sánchez +7
Medicine · Computer Science · Health Professions · #Facial Nerve Paralysis Treatment and Research #Face recognition and analysis #Temporomandibular Joint Disorders
paper · pdf · doi:10.48550/arxiv.2004.06657
Action Units (AUs) are geometrically-based atomic facial muscle movements\nknown to produce appearance changes at specific facial locations. Motivated by\nthis observation we propose a novel AU modelling problem that consists of\njointly estimating their localisation and intensity. To this end, we propose a\nsimple yet efficient approach based on Heatmap Regression that merges both\nproblems into a single task. A Heatmap models whether an AU occurs or not at a\ngiven spatial location. To accommodate the joint modelling of AUs intensity, we\npropose variable size heatmaps, with their amplitude and size varying according\nto the labelled intensity. Using Heatmap Regression, we can inherit from the\nprogress recently witnessed in facial landmark localisation. Building upon the\nsimilarities between both problems, we devise a transfer learning approach\nwhere we exploit the knowledge of a network trained on large-scale facial\nlandmark datasets. In particular, we explore different alternatives for\ntransfer learning through a) fine-tuning, b) adaptation layers, c) attention\nmaps, and d) reparametrisation. Our approach effectively inherits the rich\nfacial features produced by a strong face alignment network, with minimal extra\ncomputational cost. We empirically validate that our system sets a new\nstate-of-the-art on three popular datasets, namely BP4D, DISFA, and FERA2017.\n