Supervisor: Shan Du
Student: Wenqi Marshall Kwok Guo
This course is a direct study focused on computer vision techniques, specifically the segmentation of deformable objects using semi-supervised methods with a small amount of labeled data. For example, the student's method should be able to segment smoke or colored gas from an image with fewer than 10 human-labeled images. The student is expected to conduct rigorous experimentation to evaluate the performance of the proposed method against existing benchmarks. The course will culminate in a comprehensive analysis of the method's characteristics and potential applications. Weekly meetings with the supervisor will guide the research, and the student is expected to deliver a final report and presentation summarizing the findings.
| Evaluation Component | Deadline | Grade Weight |
|---|---|---|
| Written Literature Review | September 20 | 10% |
| Weekly Meeting and Deliverables | Ongoing, Weekly | 10% |
| Custom Model Development | November 1 | 25% |
| Baseline Model Comparisons | November 30 | 15% |
| Final Report | Last Day of Classes | 20% |
| Final Presentation | Last Day of Classes | 20% |
The student is expected to dedicate approximately 8 hours per week to course-related activities, including research, experimentation, and meetings with the supervisor.