Jun. 2024 - Feb. 2025
Learning and Vision Lab, ECE Dept., National University of Singapore
A primarily feed-forward framework for dynamic scene reconstruction. At its core, it leverages temporal correspondences to extend existing 3D formulation to 4D.
Apr. 2024 - Jun. 2024
Learning and Vision Lab, ECE Dept., National University of Singapore
A framework that leverages the explicit representation power of 3D Gaussian Splatting and conceptualizes the video content as a fluid flow of Gaussian points through space and time, effectively reconstructing a 4D world without direct camera input .
Aug. 2023 - Apr. 2024
Temasek Laboratories , National University of Singapore
Designed a union event camera drone detection working flow to process dataset and implement ordinary object detection methods on high frequency event data without changing the original network structure.
Combined three fully labeled or semi-labeled drone detection frame dataset and one unlabeled lab-collected event dataset into one fully labeled event camera dataset. 
Apr. 2023 - Apr. 2024
Learning and Vision Lab, ECE Dept., National University of Singapore
Focuse on understanding and disentangling , graspe the role of a specific region of a image or a component in the network and analyzed the decision flow of pre-trained networks and post-trained networks.
Dec. 2022 - Mar. 2023
Learning and Vision Lab, ECE Dept., National University of Singapore
National University of Singapore (Chongqing) Research Institute
Recover visual information from high resolution non-homogeneous hazy images, combine dehazing with downsample and super resolution, integrate WeatherDiffusion as the main dehaze module and DehazeFormer as the fine-tuning and super resolution module. 
Sept. 2022 - Jun. 2023
Learning and Vision Lab, ECE Dept., National University of Singapore
National University of Singapore (Chongqing) Research Institute
Machine Intelligence Lab., College of Computer Science, Sichuan University
Utilized 2D images or 2D coordinates of the human body to estimate the real-world 3D coordinates of the key points of the human body, combined weak supervision with the well-performed GCN network.
Used the parameter of bone length to constrain the self supervision learning part in the weak supervision. 
Mar. 2022 - Mar. 2023
Machine Intelligence Lab., College of Computer Science, Sichuan University
Proposed a seven-class facial expression recognition data-set with face mask, FM-FER2013.
Introduced a Facial Expression Recognition Method with Face Mask, used YOLOv5 for face detection and face mask detection, and Resnet for facial expression recognition.
May 2022 - Jul. 2022
NUS School of Computing Summer Workshop 2022
Used LSTM to online training and predicting the price of each stock, utilized Vue framework to visualize in website and interacting with user, scored A in AI/ML for Workshop Financial Service Track.
Nov. 2020 - Jul. 2022
Chengdu Yinlaiyinwang (Convenient Printing) Technology Co., LTD
Led a team of 10 and successfully developed an intelligent online printing system called Yinlaiyinwang, transforming traditional offline printers into internet-connected smart devices.
Established an on-campus experience store that served over 100,000 students and offered more than 15 part-time job opportunities in campus.
Earned multiple entrepreneurship awards at both the college and university levels.
Sept. 2020 - Jul. 2021
Student Innovation and Entrepreneurship Training Program, Sichuan University
Designed a defense mechanism using Deep K-NN to detect and filter poisoned samples during training, enhancing the robustness of face recognition systems against adversarial data injection.
Sept. 2020 - Jul. 2021
Student Innovation and Entrepreneurship Training Program, Sichuan University
Developed a remote rehabilitation assessment system using OpenPose and deep learning to evaluate patients’ movements via video, enabling accessible at-home physical therapy monitored by clinicians.