readme

README.md

@@ -18,6 +18,7 @@ library_name: diffusers
 
 ## VAE Training Process
 
+ - Encoder: Frozen (to avoid retraining SDXL for the new VAE).
  - Dataset: 100,000 PNG images
  - Training Time: 4 days
  - Hardware: Single RTX 4090
@@ -28,17 +29,16 @@ library_name: diffusers
 
 ## Implementation
 
-Base Code: Used a simple diffusion model training script.
-Encoder: Frozen (to avoid retraining SDXL for the new VAE).
-Training Target: Only the decoder, focusing on image reconstruction.
+ - Base Code: Used a simple diffusion model training script.
+ - Training Target: Only the decoder, focusing on image reconstruction.
 
 ## Loss Functions
 
-Initially used LPIPS and MSE.
-Noticed FID score improving, but images becoming blurry (FID overfits to blurry images—improving FID is not always good).
-Switched to MAE (Mean Absolute Error) instead of MSE (not sure is MSE bad).
-Balanced LPIPS and MAE at 90/10 ratio.
-Used median perceptual_loss_weight for better balance.
+ - Initially used LPIPS and MSE.
+ - Noticed FID score improving, but images becoming blurry (FID overfits to blurry images—improving FID is not always good).
+ - Switched to MAE.
+ - Balanced LPIPS and MAE at 90/10 ratio.
+ - Used median perceptual_loss_weight for better balance.
 
 ## Results