UCSD ECE 285: Introduction to Visual Learning (Spring 2022)

Final Project

There are two directions for the final project. Students can form teams with 1-2 people to finish the project.

The first direction is: You can propose a computer vision related project, and then do experiments and make a final report.

The second direction is: Find existing recent work and re-implement it. The rules are:

1. You need to reimplement the whole thing, you cannot borrow any things from the existing code.
2. You need to apply the algorithms on a different dataset that is not mentioned in the paper, you can even collect the dataset and re-define the problem yourself.

Project Proposal (10%) [Template]

Submit a proposal stating the problem, motivation, approach, and the dataset you are going to work on in a PDF of no less than 1 page.

Final Project Submission (30%) [Template]

• A final project report, extending from the proposal: stating the problem, motivation, approach, and dataset. Report the final results and analysis of the results.
• Code
• A 5-min recorded video demo introducing your method and the visualization of the results.

Topics for Second Direction

The paper you select can be from the following topics or it can be other papers related to the class, please talk to the TA and instructor for selecting the topic:

1. Generative Models (note you can find other papers under a similar topic)
   • GANs [paper]
   • Conditional GANs [paper]
   • VAEs [paper]
2. Test-Time Training with Self-Supervision for Generalization under Distribution Shifts
   • [paper]
3. Image Inpainting (note you can find other papers under a similar topic)
   • [paper-1]
   • [paper-2]
4. Surface Normal Prediction (note you can find other papers under a similar topic)
   • [paper]
5. Optical Flow Estimation (note you can find other papers under a similar topic)
   • [paper]