👤 Short Bio
Welcome to my academic homepage. I am Jiayi Wu, a Ph.D. student supervised by Prof. Yiannis Aloimonos at PRG (Perception & Robotics Group) Lab, University of Maryland, College Park. My research focuses on computer vision and robotics, specifically in the areas of active vision, field robotics, and underwater 3D vision. Inspired by nature, I am dedicated to creating innovative robot perception systems that challenge existing paradigms and redefine the future of robotics.
My research interest includes:
- General Environment Vision: Make computer vision systems more robust to the environment (scattering media, bad weather)
- Low-cost 3D reconstruction and depth estimation in scattering medium
- Active Vision: Feedback control through vision
2023/06/06: Our paper “3D Reconstruction of Underwater Scenes using Nonlinear Domain Projection” won Best Paper Award at the IEEE Conference on Artificial Intelligence (IEEE CAI) 2023, Santa Clara, California !!!
2023/05/01: Our paper “3D Reconstruction of Underwater Scenes using Nonlinear Domain Projection” has been accepted by IEEE Conference on Artificial Intelligence (IEEE CAI) 2023 !!!
2023/04/03: Our paper “VALIDATION OF A FULL-WAVE BACKSCATTER MODEL FOR CORN FIELDS USING MEASUREMENTS FROM A GROUND-BASED SCATTEROMETER” has been accepted by IGARSS 2023 !!!
2023/01/16: Our paper “UDepth: Fast Monocular Depth Estimation for Visually-guided Underwater Robots” has been accepted by ICRA 2023 !!!
2022/09/15: The paper “UDepth: Fast Monocular Depth Estimation for Visually-guided Underwater Robots” I collaborated with Boxiao Yu and Prof. Islam was submitted to ICRA 2023.
2022/08/23: Completed the summer internship at Vobile and got a return offer (recommended by Dr. Zhao).
2022/05/23: Joined the team of Dr. Zhao (CTO of Vobile) of Vobile as an audio and video algorithm development engineer (summer internship).
2022/01/10: Join Professor Judge’s Remote Sensing Lab (as Graduate Student Assistant) and be responsible for the development of a large-scale automated generation of 3D plant models.
- Ph.D. in Computer Science ——— Aug. 2023- Present
Supervised by Prof. Yiannis Aloimonos.
- M.S.(Thesis) in Electrical and Computer Engineering ——— Aug. 2021- May. 2023
Supervised by Prof. Md Jahidul Islam.
- B.E. in Mechatronic Engineering ——— Sept. 2017- Jun. 2021
2021 Outstanding Graduate, Zhejiang Sci-Tech University
Wu, Jiayi. Low-Cost Depth Estimation and 3D Reconstruction in Scattering Medium. Master’s Thesis. 2023
A. Kaleo Roberts, Kamal Sarabandi, Jasmeet Judge, Alejandro Monsivais-Huertero,
Jiayi Wu. VALIDATION OF A FULL-WAVE BACKSCATTER MODEL FOR CORN FIELDS USING MEASUREMENTS FROM A GROUND-BASED SCATTEROMETER. IGARSS. 2023
- A monocular underwater depth estimation pipeline, using RMI as the input space, constructs a lightweight domain projection module, a lightweight CNN feature extraction module, and a lightweight Transformer-based depth estimation network. And the depth estimation network is supervised with knowledge of underwater light attenuation as a prior. The results show that our depth estimation accuracy is close to the SOTA model, and much faster than them.
This is a collaborative project on underwater depth estimation with Boxiao Yu, a Ph.D. student in RoboPI lab.
[IEEE Xplore] [arXiv] [Code] [pre-print]
- A fast depth-guided semi-dense underwater 3D reconstruction pipeline without a deep learning model. Underwater image restoration is implemented through RMI channel and underwater imaging model to improve the number of feature matches. Reduce the computational complexity of feature extraction by using the depth map as a mask. In order to achieve higher quality(semi-dense) underwater 3D reconstruction, the depth maps are used as the supervision to refine the 3D point cloud and remove noise.
This project is my master’s thesis, we have submitted a journal to the IEEE Transactions on Artificial Intelligence (TAI). Here is the video demo.
Learning-based Infringing Video Retrieval
- An learning-based infringing video retrieval system based on the fusion of global features and local features. I did a summer internship in the Vobile’s R&D department under the supervision of Dr. Zhao, CTO of Vobile. I built a learning-based infringing video retrieval system based on the fusion of global features and local features. I used Vobile’s video database to train the model, and achieved good performance. Before I ended my internship, I put it together into a python package and organize each component in the system into an easy-to-use python toolkit. I also wrote the manual of the package and a document about future optimization steps to finally handed over to the person who took over the project.
🏭 Job Experience
Digital Audio and Video Algorithm Engineer ——— May. 2022- Aug. 2022
Vobile, Santa Clara, CA, United States.
Graduate Student Assistant ——— Jan. 2022- present
Remote Sensing Laboratory at University of Florida, Gainesville, FL, United States.
🏅 Honors and Awards
- 2020.12 Zhejiang Government Scholarship
- 2019.09 First Class School Financial Aid for Overseas Exchange Program
- 2018.12 Zhejiang Government Scholarship
- 2021.06 Individual
1st Prizein the National University Graduate Design Competition (Only two people won this award nationwide) 06/2021
- 2019.10 Provincial
1st Prizeof National 3D Digital Innovative Design Competition
2nd Prizeof National 3dds Competition Classic
3rd Prizeof The 16th Zhejiang Province Mechanical Design Competition for College Student
3rd Prizeof The Challenge Cup Extracurricular Academic Works Competition
💬 Academic Services
- 2023 IEEE International Conference on Robotics and Automation (ICRA)
- 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2023)