My Project

Video based Speed Measurement

The project focused on vehicle speed estimation based on videos and was therefore built based on object recognition and tracking algorithm. A calibration method is also implemented here to fix the tortion of distance caused by Perspective. Detailed code will be released with paper later.

show details >>

OCR Detection Paper Review

OCR is a nice tool, used frequently in area. It is also developing rapidly in recent years with the development of nature language processing.

show details >>

Analysis of Vaccination on Markovian Random Walk Model of epidemic proliferation

The project simulated the spread of the epidemic under SIR model by using Markovian random walk and the paper then discussed the impact of epidemic prevention policies and vaccines on the spread of the epidemic.

show details >>

Automated Spine Vertebrae Segmentation

Automated Vertebrae Segmentation provides multiple benefits including improving time efficiency, facilitating 3D modeling and supporting diagnosis (Spinal Fractures, Tumors and Metastasis, Osteoporosis, etc.). Nevertheless, training a model on the 24 vertebrae is challenging, due to inaccuracy, it is very likely disorderly labeled. Training a model on separating the spine into Cervical, Thoracic and Lumbar is much easier and more accurate on the other hand. Therefore, can we think of a method that automatically separate the vertebrae within the three parts?

show details >>

Improve AI-based Model Robustness of Ventricle Segmentation from Multi-Contrast MRI via Active Learning

Accurate and automatic ventricle segmentation is critical for diagnosing hydrocephalus and monitoring its progression in patients. However, during model training, we observed challenges in achieving robust performance across varying MRI contrasts and diverse patient populations. Active learning, known for its ability to minimize labeling costs and expedite model adaptation to different data populations, offers a promising solution to these issues. In the paper, we present a novel uncertainty-based metric designed to assess the informativeness of individual data samples. By integrating this metric into the active learning process, we aim to improve model robustness and generalization across multiple MRI contrasts, enabling more reliable ventricle segmentation even in complex clinical scenarios.

show details >>

Automatic video edition and highlighting on surgery videos

Yijing is a Chinese ancient classic masterpiece that expounds the changes of the earth. Only one of the three parts of it exists now and it explains the connections between human and the nature. It has also been used for divination...

show details >>