This project presents a simple yet effective semi-supervised learning framework for medical image segmentation. It constructs two sets of Mean Teacher models to form a co-training framework and leverages the diversity between the representational capabilities of different networks at different iterations to build stable diversity among models. Subsequently, it utilizes pseudo-label filtering and difference-driven alignment to introduce robust regularization constraints for training. The paper achieves state-of-the-art performance in the field and validates the method’s generalization ability across various datasets.

During the project, my responsibilities included:

• Independently proposing research ideas and conceptualizing model implementation.
• Independently completing experimental work for the paper (including comparative experiments and ablation studies)
• Worked closely with the supervisor to ensure the quality of paper writing, figure drawing and addressing rebuttals.

As the leader of the vision team, I organized and led the team to build the visual framework, implement algorithms, deploy, and maintain them. I undertook the main responsibilities in the project. I collaborated with the mechanical and electrical control teams and led the team to win the first prize nationwide for building a robot from scratch.

During the project, my responsibilities included:

• Detecting targets using both traditional vision methods and deep learning.
• Implementing target tracking using filtering algorithms(e.g., EKF).
• Debugging and developing SLAM programs based on ROS for autonomous navigation and obstacle avoidance.
• Development using Linux systems.