NAVER LABS Europe seminars are open to the public. This seminar is virtual and requires registration
Date: 12th May 2022, 10:00 am (GMT +2:00)
Self-supervised frameworks for 3D human pose estimation
Abstract: Available 3D human pose estimation approaches leverage different forms of strong (2D/3D pose) or weak (multi-view or depth) paired supervision. Barring synthetic or in-studio domains, acquiring such supervision for each new target environment is highly inconvenient. To this end, we cast 3D pose learning as a self-supervised adaptation problem that aims to transfer the task knowledge from a labeled source domain to a completely unlabeled target. The main challenge is determining proper learning objectives in order to obtain supervision from the data itself. In one of our works, the key contribution is the effective use of new inter-entity relationships to discern the co-salient foreground appearance and thereby the corresponding pose from just a pair of images having diverse backgrounds. While in another work, we achieve the same by distilling well-formed latent non-local relations in the pose and motion space. Further, in a recent work, we propose a novel way to perform self-adaptive 3D human pose estimation by enabling uncertainty estimation. We develop learning techniques to make the model behave differently for the in-domain and out-of-domain scenarios. Later, the same instigated behavior is used to devise effective adaptation objectives. The ability to discern out-of-domain samples allows a model to assess when to perform re-adaptation while deployed in a continually changing environment. Such solutions are in high demand for enabling effective real-world deployment across various industries, from virtual and augmented reality to gaming and healthcare applications.
About the speaker: Jogendra is wrapping up his PhD at Video Analytics Lab, CDS, Indian Institute of Science. He is broadly interested in self-supervised learning, domain adaptation, and their application to computer vision problems like 3D human pose estimation, semantic segmentation, etc. He is a recipient of several PhD fellowships, like Wipro PhD fellowship, Qualcomm Innovation Fellowship, and the Google PhD fellowship. He has worked as a Research Intern at Facebook Reality Labs during Fall 2021. He has authored papers in several top-tier venues such as CVPR, ICCV, ECCV, NeurIPS, AAAI, and WACV. He has also served as a reviewer in several top-tier conferences and journals (T-PAMI and IJCV) and has received Outstanding Reviewer awards at CVPR 2020 and NeurIPS 2021.