Dongyu Yan 💫 (闫栋宇)

I'm a PhD student (2024 ~ 2028), working with Prof. Ying-Cong Chen at AI Thrust, Information Hub, Hong Kong University of Science and Technology (Guangzhou).

I receievd my Master degree (2021 ~ 2024) in the School of Mechanical Engineering and Automation at Harbin Institute of Technology (Shenzhen), supervised by Prof. Haoyao Chen in nROS-lab. I obtained a B.Eng (2017 ~ 2021) from Harbin Institute of Technology.

I was a research intern at the Department of Flight System of DJI (Dec. 2021 ~ Oct. 2022). Before that, I was a research intern at the Department of Transformer of Megvii (May. 2021 ~ Aug. 2021).

Email  /  Scholar  /  Github  /  Twitter  /  Bilibili

Research

My research interest lies in computer vision, 3D generation, and robotics. I used to work on topics that combine implicit 3D representation with robotics tasks, including neural reconstruction, neural SLAM, and implicit NBV planning. Recently, I'm interested in 3D diffusion models, especially in the context of object texture generation and scene generation. My research object is to build efficient and powerful 3D generative models that can be really applied by industries like game asset creation, special effect production and computer aided designing. Below are some of my selected papers. Some papers are highlighted.

Selected Publications

MSI-NeRF: Linking Omni-Depth with View Synthesis through Multi-Sphere Image aided Generalizable Neural Radiance Field
Dongyu Yan, Guanyu Huang, Fengyu Quan, Haoyao Chen.
IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
Arxiv / Paper / Video / Github

We create a generalizable NeRF that takes four fisheye images as input and outputs a 3D MSI representation for novel view synthesis and depth estimation. It can be trained with synthetic depth data only and can generalize to a wide range of scenarios. We also released a fisheye multi-view dataset for training and evaluation.

Active Implicit Object Reconstruction using Uncertainty-guided Next-Best-View Optimization
Dongyu Yan*, Jianheng Liu*, Fengyu Quan, Haoyao Chen.
IEEE Robotics and Automation Letters (RA-L)
Arxiv / Paper / Video / Github

We propose an active implicit object reconstruction method leveraging direct uncertainty evaluation from implicit occupancy map and Next-Best-View optimization. It directly optimizes a vitrual camera trajectory in the uncertainty field to maximize the information gain for the current reconstruction.

Efficient Implicit Neural Reconstruction Using LiDAR
Dongyu Yan, Xiaoyang Lyu, Jieqi Shi, Yi Lin.
IEEE International Conference on Robotics and Automation (ICRA)
Arxiv / Paper / Video / Github / Project Page

We propose an implicit reconstruction method that uses LiDAR scans as input, which is efficient, accurate, and can be applied in various scenarios. This method can be applied for real-world scenes, even with sparse LiDAR scans and badly aligned poses, which can be shown in the self-collected dataset we released.

Competitions

RoboMaster 2021: Ranked 2^ed
As a member of the electronic control team, responsible for the control system of the Sentry robot. Also as a robot operator.
Robocon 2020 China Division: Ranked 3^rd
As a leader of the electronic control team, responsible for the development of the PMSM Driver system.
RoboMaster 2019: Ranked 5-6^th
As a leader of the computer vision team, responsible for the development of the SLAM and Navigation system.

My time in HITCRT taught me a lot. Because of it, I embarked on the path of scientific research today. I will always be grateful to my teammates who fought alongside me. Hope our team can gets better and better.

About Me

Maniac of Racing and Car Modification
I love to free myself on the race track. I have a Honda FIT which is well modified by myself and I hope to take it to every famous track in the world. I'm also interested in sim racing and have a cockpit in my dormitory to practice. I'm currently participating in Hipole's sim racing championship

Big Fan of Music
I'm a big fan of music. My favourite music genre is Vocaloid and Japanese Pop. I can play violin and now I'm learning to play guitar. I'm also a begginer of music composing and I hope to become a Vocaloid producer in the future.

Standard Otaku
Since 2010s, I started to watch anime. My Character Name (CN) is 星空ほしそら. My favourite nijigen character is Hatsune Miku, a famous virtual singer from Japan. I also have an Miku Itasha (the Honda FIT mentioned before), which makes me a member of one of the biggest Itasha group in China called Hatsune Miku Itasha Lab.

Skills
Python   /   C++   /   Matlab   /   Linux   /   Embedded System
CAD   /   PCB Design   /   Fluent English & 日本語

「愛されなくてもいいよ、君がいるなら！」

Great thanks to Jon Barron for providing the website source code.