Yan Hao

I'm Yan Hao, a computer science master student at ETH Zürich. Before that, I received the B.S. degree at ACM Honors Class, Zhiyuan College, Shanghai Jiao Tong University (SJTU).

I am interested in machine learning and computer vision, especially 3D vision. At ETH, I contributed to a 3D Vision project supervised by Prof. Dr. Iro Armeni. My master thesis was done at EPFL under the supervision of Dr. Florent Forest and Prof. Dr. Olga Fink.

Email  /  Github  /  Google Scholar  /  CV

Education
Master Thesis, EPFL, Switzerland
Nov. 2022 - June. 2023
Master in Computer Science, ETH Zürich, Switzerland
Sep. 2020 - Present
B.S. in Computer Science, Shanghai Jiao Tong University
ACM Honors Class, supervised by Prof. Yong Yu.
Sept. 2016 to Jun. 2020
Internship
Research Intern, Amazon AWA Shanghai AI Lab
Supervised by Tong He and Tianjun Xiao.
Jun. 2020 to Sept. 2020
Research
Source Free Domain Adaptation for Object Detection applied to Road Scene Understanding
Yan Hao*, Florent Forest*, Olga Fink
We propose a new source-free domain adaptation method for object detection applied to road scene understanding. Implementation bases on Detectron2 and Pytorch. Achieve state-of-the-art results for the adaptation from Cityscapes to Cityscapes Foggy and from Sim10k to Cityscapes.
Master Thesis. In preparation for conference submission.
Nothing Stands Still: A Spatiotemporal Benchmark on 3D Point Cloud Registration Under Large Geometric and Temporal Change
Tao Sun, Yan Hao, Shengyu Huang, Silvio Savarese, Konrad Schindler, Marc Pollefeys, Iro Armeni
We propose a new spatiotemporal dataset and benchmark called NSS (Nothing Stands Still) on 3D point cloud registration under large geometric change across temporal stages. Accepted at CVPR 2023 DEMO.
In submission.
3D Objectness Estimation via Bottom-up Regret Grouping
Zelin Ye, Yan Hao, Liang Xu, Rui Zhu, Cewu Lu
We propose a robust 3D objectness estimation method in a bottom-up manner, i.e. first over-segment scene pointclouds and then group them iteratively with a novel regret mechanism to withdraw incorrect groupings.
Arvix.
PAL-Net: Predicate-Aware Learning for Scene Graph Generation
Liang Xu, Yong-Lu Li, Minyang Chen, Yan Hao, Cewu Lu
Our proposed PAL-Net has two ingredients for scene graph generation. First we introduce a novel embedding loss for translation embedding in a metric learning manner. Then we take predicates as conditions for contextualmodeling to alleviate noise.
ICME. Oral.
Visual Rhythm Prediction with Feature-Aligned Network
Yutong Xie, Haiyang Wang, Yan Hao, Zihao Xu
The paper proposed a data-driven visual rhythm prediction method, in which several visual features are extractedand then fed into an end-to-end neural network to predict the visual onsets.
MVA