A three day workshop on 3D Computer Vision was conducted on campus from 21 – 23 February. The domain of 3D Vision deals with acquisition and analysis of 3-dimensional objects/scenes with wider applications in Animation, AR/VR platforms, autonomous driving, medical imaging, etc. The primary representation for 3D objects are point cloud, mesh and voxels. These 3D objects/scenes are primarily captured either by depth sensors (e.g., RGBD sensors like Kinect and Laser sensors) or by processing images/videos taken from single/multiple RGB cameras. Algorithms for recovering 3D objects from multiple view images involves elaborate theoretical foundation from Multi-View Geometry developed in the last decade. Recently, deep learning methods have also reinvigorated interest in this domain with more efficient/scalable solutions for 3D reconstruction and analysis.
In this course introduction to fundamentals of 3D computer vision were given with theoretical grounding to relevant methods as well as practical hands-on experience. And also an overview of recent deep learning solutions in this domain for 3D reconstruction and scene understanding were given.
The various topics covered during the workshop are introduction to imaging, sensing 3D world, stereo reconstruction, hands on session (stereo), multi-view geometry, structure from motion, expert lecture on 360* stereo, demo (stereo system Dreamvu), hands on session (SFM), deep monocular reconstruction, demo (3D body reconstruction), expert lecture on depth estimation and there was also a keynote talk on recent advancement in 3D Vision.
The panel of faculty included Prof. P J Narayanan, Dr. Anoop M Namboodiri and Dr. Avinash Sharma.