Songfang Han

Ph.D.

News

  • Jun, 2019 - Has successfully defended my Ph.D. thesis!

Intro

I have obtained my Ph.D. degree at HKUST, advised by Prof. Pedro Sander. During the Spring 2019, I was a visiting student to the University of California, San Diego, supervised by Prof. Hao Su. For the autumn 2017, I was a research intern at SenseTime, supervised by Dr. Qiong Yan. Before that, I received my B.S. degree in EE department from Tongji University in 2013.

Since April 2020, I have started at SULab as a postdoctoral researcher, working with Prof. Hao Su.

My research interest includes computer graphics and computer vision, especially in 3D geometry processing, 3D reconstruction. My work covers efficient rendering, multi-view reconstruction, portrait relighting and virtual reality.

Research

2021

Learning with Noisy Labels for Robust Point Cloud Segmentation
Shuquan Ye, Dongdong Chen, Songfang Han, Jing Liao.
International Conference on Computer Vision (ICCV Oral), 2021
[project (code)] [paper] [code]

M3D-VTON: A Monocular-to-3D Virtual Try-On Network
Fuwei Zhao, Zhenyu Xie, Michael Kampffmeyer, Haoye Dong, Songfang Han, Tianxiang Zheng, Tao Zhang, Xiaodan Liang.
International Conference on Computer Vision (ICCV), 2021.
[paper] [code]

Compositionally Generalizable 3D Structure Prediction
Songfang Han, Jiayuan Gu, Kaichun Mo, Li Yi, Siyu Hu, Xuejin Chen, Hao Su.
arXiv, 2021.
[paper] [code] [video]

Arbitrary-Scale Upsampling Network for Point Cloud
Shuquan Ye, Dongdong Chen, Songfang Han, Ziyu Wan, Jing Liao.
Visualization and Computer Graphics (TVCG), 2021.
[project (code)] [paper]

2020

Visibility-Aware Point-Based Multi-View Stereo Network
Rui Chen, Songfang Han, Jing Xu.
TPAMI 2020 Apr 22. doi: 10.1109/TPAMI.2020.2988729.
[project (code)] [paper]
This is an extension of our ICCV work (Point-based Multi-view Stereo Network). In this paper, we introduce visibility-aware multi-view feature aggregation modules to gather information from visible views only for better depth prediction accuracy.

2019

Point-Based Multi-View Network
Songfang Han*, Rui Chen*, Jing Xu, Hao Su.
ICCV 2019 (oral).
[project (code)] [paper] [video]
An iterative framework to predict the depth of a scene using point cloud representation from multiple views. Use deep learning over the kNN graph to predict the residual for geometry estimation refinement.

2018

In-Depth Buffers
Songfang Han, Ge Chen, Diego Nehab, Pedro Sander.
Proceedings of the ACM on Computer Graphics and Interactive Techniques (PACM) 2018.
[paper] [code] [video]

2017

Triangle Reordering for Efficient Rendering in Complex Scenes
Songfang Han, Pedro Sander.
The Journal of Computer Graphics Techniques (JCGT) 2017.
[paper] [code] [video]

2016

Triangle Reordering for Reduced Overdraw in Animated Scenes
Songfang Han, Pedro Sander.
Proceedings of the 20th ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games (I3D) 2016.
[paper]