Workshop on Learning to See from 3D Data
in conjunction with ICCV2017
Sala Perla (Palazzo del Casino)
Recently, large-scale repositories such as ShapeNet have made available 3D geometry data with properties such as texture, object part annotations, orientation alignment, and object semantics.
The geometric and material information in these repositories has catalyzed machine learning techniques for understanding the real world. In particular, during the past few years, we have witnessed pioneering work published at ICCV, ECCV, CVPR, and NIPS that leverages learning from big 3D data, on a broad range of topics including 3D reconstruction from single images, novel-view synthesis, 3D shape space learning, 3D shape completion, and joint 2D image and 3D shape analysis.
We propose to organize this workshop in order to coordinate the effort of the research community on tackling these problems. The workshop is comprised of two parts. One part will present results from an upcoming ShapeNet-related challenge on Large-scale 3D Understanding and Synthesis Challenges. The other part will consist of invited talks by leading researchers on various topics related to learning from big 3D data, especially synthetic data. Next, we discuss the plans for both parts in details.
Large-Scale 3D Shape Reconstruction and Segmentation from ShapeNet Core55, Li Yi, Hao Su, Lin Shao, Manolis Savva, Haibin Huang, Evangelos Kalogerakis, Thomas Funkhouser, Leonidas Guibas, et al. PDF
This report summarizes the results from ShapeNet challenge, including an overview of methods from participating teams, their performances, and discussions.