CVPR 2020 Workshop
Frontiers of Monocular 3D Perception
Toyota Technological Institute at Chicago (TTIC) & Toyota Research Institute (TRI)
June 2020, Seattle, USA
This workshop will consider challenges and opportunities at the frontier of 3D reasoning from monocular visual input, the main modality common to most mobile phones, robots, and cars. Our main focus will be on 3D perception of scenes and objects directly from images and videos. This includes key challenges in Computer Vision like depth estimation, 3D object detection, and reconstruction. Exciting new work is being done in this area, including significant improvements on existing benchmarks thanks to novel CNN-based and self-supervised approaches that effectively combine the representational capacity of deep learning with geometry.
The proposed workshop’s main theme will be to identify, characterize, and investigate how to qualitatively go beyond the current limitations in monocular 3D perception. We will discuss and debate the fundamental goals of depth perception, the merits of new data (e.g., DIODE; MatterPort 3D; and 3D movies), different learning paradigms (e.g., purely supervised with LiDAR and/or Kinect-type devices; self-supervised using videos; and semi-supervised), new metrics, and new tasks. Our goal is to use this workshop to disseminate state-of-the-art technical knowledge on 3D perception, use the proposed benchmark (see below) to assess where we are, and to provide a venue for meaningful and vigorous debate of current challenges and ways to overcome them. We hope that the workshop will result in a consensus among multiple groups working in the area on how to effectively and synergistically pool data collection and evaluation efforts across the community.
Call for Papers