20160831-29 苏昊：3D from Single Images using a Large-scale Shape Repository

报告嘉宾1：苏昊 （斯坦福大学）

报告时间：2016年8月31日（星期三）晚20:00（北京时间）

报告题目：3D from Single Images using a Large-scale Shape Repository

主持人：任传贤（中山大学）

报告摘要：

Recent technological developments have led to an explosion in the amount of 3D data. With this background, we build a knowledge-base that stores rich information on top of networks of 3D models. Knowledge stored on ShapeNet can be used as priors of geometry and physics attributes for understanding contents of RGB(D) images.

We believe that such a 3D-centric knowledge-base, named ShapeNet, is beneficial to many aspects of our life, since we live in a world where 3D is the natural space. We jointly analyze 3D models and 2D images for various applications. The key idea is to bridge CG and CV: we render ShapeNet models into large volume of images with free yet detailed annotation. With such synthetic data in the virtual environment, we train high capacity models such as CNN and adapt the learned models in real world scenarios. We show various applications that achieve state-of-the-art performance, such as single-image based 3D reconstruction, image-based shape retrieval, 3D viewpoint estimation, texture transfer from images to shapes, 3D human pose estimation, and novel view feature synthesis.

参考文献：

[1] Angel X. Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva*, Shuran Song, Hao Su*, Jianxiong Xiao, Li Yi, and Fisher Yu, “ShapeNet: An Information-Rich 3D Model Repository”, arxiv, correspondence author

[2] Hao Su*, Charles Qi*, Matthias Niessner, Angela Dai, Mengyuan Yan, Leonidas Guibas, “Volumetric and Multi-View CNNs for Object Classification on 3D Data”, CVPR 2016 (spotlight)

[3] Hao Su*, Charles Qi*, Yangyan Li, Leonidas Guibas, “Render for CNN: Viewpoint Estimation in Images Using CNNs Trained with Rendered 3D Model Views”, ICCV 2015 (oral)

[4] Hao Su*, Fan Wang*, Li Yi, Leonidas Guibas, “3D-Assisted Image Feature Synthesis for Novel Views of an Object”, ICCV 2015 (oral)

[5] Hao Su*, Yangyan Li*, Charles Qi, Noa Fish, Daniel Cohen-Or, Leonidas Guibas

“Joint Embeddings of Shapes and Images via CNN Image Purification”, SIGGRAPH Asia 2015

[6] Hao Su, Qixing Huang, Niloy Mitra, Yangyan Li, Leonidas Guibas, “Estimating Image Depth using Shape Collections”, SIGGRAPH 2014

报告人简介：

Hao Su is currently a Ph.D candidate in the Computer Science Department of Stanford University. He is a member of Stanford AI Lab and Geometric Computing Lab. He served as the chair of CVPR’15 workshop on 3D from single images, ICCV’15 workshop on 3D representation and recognition, ECCV’16 workshop on augmented reality and virtual reality, and is the publication chair of International Conference on 3DVision 2016. He is also a keynote speaker at NIPS’16 workshop on 3D deep learning.

Hao’s research interest includes computer vision, computer graphics and machine learning. He has published papers at CVPR, ICCV, NIPS, ICML, SIGGRAPH, SIGGRAPH Asia, VLDB, SIGSPATIAL, IJCV, etc. He is currently a student lead of the ShapeNet team and used to be a student lead of the ImageNet team.

He is co-advised by Prof. Leonidas Guibas and Prof. Silvio Savarese. He was formally advised by Prof. Wei Li (Beihang), Prof. Fei-Fei Li (Princeton & Stanford), and Dr. Harry Shum (Microsoft). He has also been conferred a Ph.D degree in Mathematics at Beihang University.