M2 Internship : Human pose estimation
2 Mars 2022
Catégorie : Stagiaire
Context and goals.
Human pose estimation has always been an active research domain in computer vision. Early solutions to this problem are based heavily on classical machine learning. Thanks to the extensive development of recent deep learning methods that revolutionize many domains in computer vision and the availability of large-scale datasets of human pose, many issues in human pose estimation have been addressed. In many real-world applications, however, existing human pose estimation methods are not mature enough to build trustworthy AI that can replace humans. Some examples are movement assessment, analysis of injuries, and action recognition in sports, to name a few. These applications require real-time, highly accurate human pose estimation techniques that are able to deal with fast movements, unusual poses, self-occlusion, etc. Recently, Graph Convolutional Networks (GCNs) [1, 2, 3, 4, 5] have increasingly attracted attention in the machine learning community and these models have demonstrated state-of-the-art performance in many machine learning tasks. GCNs generalize many concepts of Convolutional Neural Networks to deal with graph data. Although a variety of GCNs has been proposed to address 3D human pose classification, their application in 3D human pose estimation is currently underexploited. GCNs are particularly well suited for analysing human pose data since they can capture complex dependencies of human joints during an action where the human skeleton can be naturally represented by a graph. The goal of this internship is therefore to study GCNs for 3D human pose estimation.
Profile and requirements.
The knowledge needed for this internship includes some background in machine learning/computer vision. Experience in deep learning frameworks such as Pytorch/Tensorflow will be appreciated.
Location and Period.
This internship will take place at ENSEA in Cergy, France, and Vrije Universiteit Brussel in Bruxelles, Belgique. The “gratification de stage” (compensation) is approximately 550e/month. The duration of the internship is 5-6 months. The intership is expected to start in february/march 2022 but the precise starting and ending dates can be adjusted according to the availability of the selected candidate.
Please send a CV and short statement of interest to email@example.com; firstname.lastname@example.org; email@example.com.
 Ke Cheng, Yifan Zhang, Xiangyu He, Weihan Chen, Jian Cheng, and Hanqing Lu. Skeleton-Based Action Recognition With Shift Graph Convolutional Network. In CVPR, pages 180-189, 2020.
 Michael Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering. In NIPS, pages 3844-3852, 2016.
 Martin Simonovsky and Nikos Komodakis. Dynamic Edge-Conditioned Filters in Convolutional Neural Networks on Graphs. CoRR, abs/1704.02901, 2017.
 Sijie Yan, Yuanjun Xiong, and Dahua Lin. Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition. In AAAI, pages 7444-7452, 2018.
 Xikun Zhang, Chang Xu, Xinmei Tian, and Dacheng Tao. Graph Edge Convolutional Neural Networks for Skeleton Based Action Recognition. CoRR, abs/1805.06184, 2018.