GAMES Webinar 2020 – 166期(视觉与成像专题) | Yi Zhou (Adobe Research), Jun Gao (University of Toronto)
【GAMES Webinar 2020-166期】(视觉与成像专题)
报告嘉宾1： Yi Zhou (Adobe Research)
报告题目：Fully Convolutional Mesh Autoencoder using Efficient Spatially Varying Kernels
Learning latent representations of registered meshes is useful for many 3D tasks. Techniques have recently shifted to neural mesh autoencoders. Although they demonstrate higher precision than traditional methods, they remain unable to capture fine-grained deformations. Furthermore, these methods can only be applied to a template-specific surface mesh, and is not applicable to more general meshes, like tetrahedrons and non-manifold meshes. While more general graph convolution methods can be employed, they lack performance in reconstruction precision and require higher memory usage. In this talk, I will introduce a non-template-specific fully convolutional mesh autoencoder for arbitrary registered mesh data. It is enabled by our novel convolution and (un)pooling operators learned with globally shared weights and locally varying coefficients which can efficiently capture the spatially varying contents presented by irregular mesh connections. Our model outperforms state-of-the-art methods on reconstruction accuracy. In addition, the latent codes of our network are fully localized thanks to the fully convolutional structure, and thus have much higher interpolation capability than many traditional 3D mesh generation models.
Yi joined Adobe as a research scientist in 2020. She graduated from Shanghai Jiao Tong University with a bachelor’s degree and a master’s degree in Software Engineering, and received her PhD in Computer Science from the University of Southern California, under the supervision of Dr. Hao Li and. Her research focuses on human digitizing and autonomous 3D avatars. She mainly works on deep representation learning for 3D structures like human hair, body and skeleton, and AI-driven motion synthesis and simulation.
报告嘉宾2：Jun Gao (University of Toronto)
报告题目：Learning Geometric Representation for Computer Vision
In modern computer vision, ranging from 2D image understanding to 3D object reconstruction, representation of the image or object, serving as both the input and output of a neural network, plays a crucial role for designing the network architecture and algorithms for downstream tasks. 2D images are typically processed in a discretized and fixed grid in deep CNN, 3D objects are mostly processed in the form of discretized voxels, point clouds, template mesh or implicit functions. In this talk, I argue for a more efficient and compact representation using geometric and deformable grids for 2D images and 3D objects. First, I will start by discussing the approach to use 1D spline curve to represent an MNIST image or an object mask. Then I will introduce how to design the 2D deformable grid for real-world image. At last, I will present DefTet on representing objects with arbitrary topology and geometric details in 3D. I will conclude my talk with some personal outlook on several promising directions for future research.
Jun Gao is a PhD. student in the Machine Learning Group at the University of Toronto, supervised by Professor Sanja Fidler. Jun is also a research scientist at Nvidia’s Toronto AI Lab. Jun’s research interests are broadly in deep learning, with the goal being structured and geometric representation learning, while taking insights from human’s perception on 2D images, 3D shapes and videos. Jun graduated from Peking University in 2018 with Bachelor degree, where he was fortunate to work with Professor Liwei Wang. Jun also interned in Stanford, MSRA and Nvidia.
韩晓光博士现为香港中文大学(深圳)助理教授。其研究方向包括计算机视觉、计算机图形学以及医疗图像处理等，在该方向著名国际期刊和会议发表论文近40篇，包括顶级会议和期刊SIGGRAPH, CVPR, ICCV ,ECCV, NeurIPS, AAAI, ACMMM, ACM TOG, IEEE TVCG等，其中CVPR/ICCV/ECCV三大计算机视觉顶级会议口头报告多篇。他的团队连续两年获得CVPR最佳论文提名（入选率为0.8%和0.4%）,他的团队主推的DeepFashion3D数据集获得Chinagraph开源数据集奖，他的工作曾获得计算机图形学顶级会议Siggraph Asia 2013新兴技术最佳演示奖，入选2016年年度最佳计算论文之一，他的团队于2018年11月获得IEEE ICDM 全球气象挑战赛冠军（参赛队伍1700多）。更多细节详见http://mypage.cuhk.edu.cn/academics/hanxiaoguang/
GAMES主页的“使用教程”中有 “如何观看GAMES Webinar直播？”及“如何加入GAMES微信群？”的信息；