Geometry-Aware Neural 3D Representations and Applications

报告题目：Geometry-Aware Neural 3D Representations and Applications

报告摘要：Nowadays, the ability to understand and analyze 3D data is becoming increasingly important in computer vision and computer graphics communities. Recently, 3D deep learning methods are widely used in shape classification, segmentation, and generation, etc. But most of existing methods are still constrained by the representation power of the shape descriptors due to the irregularity and sparsity of their geometric structures (e.g., 3D points and meshes). In this talk, I will introduce some of my recent research work, such as how to effectively and efficiently define the geometry-aware convolution and autoencoder on 3D point clouds, how to define a unified representation for multimodal 3D shape data (e.g., 3D point clouds and 2D images) in a joint latent space, etc. These methods can be applied in object classification, part segmentation, and semantic segmentation of large-scale scenes, as well as shape-image joint generation tasks for creating and augmenting new multimodality datasets.

讲者简介：Artem Komarichev，现博士就读于美国韦恩州立大学。主要研究方向是三维计算机视觉和几何深度学习，研究成果主要发表于CVPR，GMP,CAGD (journal of Computer Aided Geometric Design)，ICPR等，获韦恩州立大学计算机系杰出研究助理奖、杰出研究助教奖等。

时间：2022年11月28日 (周一) 上午9:00-10:00

腾讯会议：621633915