Human Faces Generation with Diffusion Models

graph TD;
	A --> B;
	A --> C;
	E --> D;
	E --> G;
	A --> F;
  A["UNet Improvements (Frank + Emre)"];
  B["Hyperparameters (Emre + Ziyu)"];
  C["Loss Functions (Ziyu)"];
  D["Diffusion Transformers (Frank + Liang)"];
  E["Latent Diffusion Models (Liang)"];
  F["Conditioning (Frank)"];
  G["Recreating Stable Diffusion (Liang + Frank)"];

Task 1 Workflow
Task 4 Workflow
1. Architecture ✅
2. Hyperparameters ✅
3. Conditioning ✅
4. Variations in training losses ✅
5. Recent variants of diffusion models ✅

🧪 Experiments

<aside> ❓

How Do I Run Ablation Experiments?

Design your experiments by holding all parameters fixed except for one variable that you change at a time. Running these experiments unfortunately means that later tasks may depend on previous tasks. However, once a task is completed, teammates can work on the sub-tasks in parallel. The experiments are structured around the 4 tasks outlined in the email.

Recommended Experiment Setup
Final Run #1
Final Run #2 </aside>

Human Faces Experiments

📚 Literature Review

Please read the papers pertaining to your tasks, especially the ones marked as Important .

🤹 Assigned Tasks

🧪 Experiments

📚 Literature Review

📒 Meeting Notes

7 May 2025

2 May 2025

2 April 2025