CS492(D): Diffusion Models and Their Applications
Minhyuk Sung, KAIST, Fall 2024
Project
Project Proposal Due: Due Oct 19 (Sat)
Project Interim Report Due: Due Nov 9 (Sat)
Project Submission Due: Due Nov 30 (Sat)
Where to submit: Gradescope
The course project is to create your own diffusion model using one of the provided datasets and benchmark setups. Below are the candidate benchmarks.
Datasets & Benchmarks (Work in progress; subject to change)
- Modeling the Distribution of Natural Disasters:
https://github.com/KAIST-Visual-AI-Group/Diffusion-Project-Natural-Disasters - Sketch Stroke Generation:
https://github.com/KAIST-Visual-AI-Group/Diffusion-Project-Drawing - Lighting-Conditioned Image Translation:
https://github.com/KAIST-Visual-AI-Group/Diffusion-Project-Illumination - 3D Volume Generation:
https://github.com/KAIST-Visual-AI-Group/Diffusion-Project-3DVolume
You’ll need to form a team with your classmates to work on the project throughout the semester, and each team must select one of the datasets/benchmarks listed above.
Project Team Matching
You are allowed to form a team with a maximum of three people. All teams will be assessed in the same way, regardless of how many people are in the team. It is hence recommended to have three people in the team. All members of a project team will receive the same grade for the project, unless under extraordinary circumstances.
Project Proposal
Submit a project proposal for each team, including the following information within three A4 pages (no template provided):
- Names and student IDs of all team members
- Dataset/Benchmark title
- Basic ideas for the diffusion model implementation
- Timeline for implementation
- Plans for task allocation to team members
Project Interim Report
Based on the timeline in the proposal, submit a write-up of up to three pages for each team, color-coding each item in the timeline as follows:
- Completed
- In progress (as scheduled)
- Delayed - Briefly explain why.
- Have issues - Briefly explain the issues.
For the cases marked as Delayed or Have issues, provide a brief explanation of the problems and propose solutions.
Keep it concise; avoid creating a lengthy write-up.
Project Submission
Submit the followings:
- A poster
- A report
- Source code and data
More details will be provided soon.
Plagiarism Policy
You are allowed to use any existing code, provided you properly cite these resources in your write-up and code. Missing references to any code, models, or data will be considered plagiarism. You are NOT allowed to use any existing pretrained models or additional data. The trained models must be reproducible with the provided datasets.