GAMES Webinar 2023 – 291期(3D Shape Modeling meets Generative Technique) | 张彪(KAUST),郑欣阳(清华大学)

【GAMES Webinar 2023-291期】(几何专题-3D Shape Modeling meets Generative Technique)



报告题目:Efficient Representations for 3D Generative Modeling


This presentation delves into the world of 3D generation models, with a spotlight on the critical significance of representations and data structures. We will survey a range of representation methods and data structures across voxel-based, mesh-based, point cloud-based, and implicit function-based domains, shedding light on their respective merits and applications. Alongside addressing the delicate balance between accuracy and efficiency, I will also share my own contributions, demonstrating how my work intersects with these foundational components. This talk will include our publications 3DILG in NeurIPS 2022 and 3DShape2VecSet in SIGGRAPH 2023.


Biao Zhang is a researcher with expertise in machine learning, deep learning, and 3D computer vision. He holds a Ph.D. from KAUST and Master’s and Bachelor’s degrees from Xi’an Jiaotong University. His research interests include 3D representation and 3D generative models, and he has published in CVPR, ICLR, NeurIPS and SIGGRAPH.





报告题目:SDF fields meet 3D shape generation


Easily creating 3D shapes to fit human’s fabulous imaginations and match the designer’s creative ideas is one of the ultimate goals in computer graphics. The rapid development of generative neural networks, such as generative adversarial networks, diffusion models, autoregressive networks, and flow-based models, achieves great progress in text, image, and video generation. These techniques have been adopted for generating 3D shapes with different kinds of 3D representations and greatly reduce the workload of 3D generation. However, there exists a large quality gap between the synthesized shapes and the dataset the generator was trained on. Moreover, existing approaches lack intuitive control and convenient ways to control the shape generation process to satisfy users’ intentions. This talk will cover two works, the first is SDF-StyleGAN, a StyleGAN2-based deep learning approach for 3D shape generation, with the aim of reducing visual and geometric dissimilarity between synthesized shapes and the shape collection, and the second is LAS-Diffusion, a diffusion-based 3D generation framework to model plausible 3D shapes via 2D sketch image input.


Xinyang Zheng is a fourth-year Ph.D. candidate of the Institute for Advanced Study, Tsinghua University, supervised by Dr. Heung-Yeung Shum. Currently, He is a research intern at the Internet Graphics group, Microsoft Research Asia, led by Dr. Xin Tong. His research interest includes 3D geometry modeling and 3D generation. His work has been published at SIGGRAPH and SGP. Before that, he received his B.S. Degree in Physics from Chu Kochen Honors College, Zhejiang University in 2020.




孟敏,广东工业大学计算机学院副教授,硕士生导师,广东省“珠江人才计划”青年拔尖人才。2011年在浙江大学获得博士学位,2011-2013年在新加坡南洋理工大学从事博士后研究。CCF计算机辅助设计与图形学专委会委员、CSIG智能图形专委会委员、三维视觉专委会委员。研究方向为图像处理、机器学习、计算机图形学。发表国际期刊和会议论文40余篇,以第一或通讯作者在CCF A类和中科院1区IEEE会刊 IEEE TIP、IEEE TCBY、IEEE TMM、IEEE TCSVT等发表论文10余篇。


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;

You may also like...