Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
| 1 |
Semantic VAE
|
| 2 |
|
| 3 |
Purpose: Extremely fast convergence on illustraton data for a downstream DiT model
|
|
@@ -14,4 +17,4 @@ VA-VAE convolution decoder
|
|
| 14 |
|
| 15 |
Loss is all layers of DINOv2 and DINOv3 equally weighted MSE on input vs reconstruction.
|
| 16 |
|
| 17 |
-
There are no pixel losses/LPIPS/GAN used, this is DINO space loss only.
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
---
|
| 4 |
Semantic VAE
|
| 5 |
|
| 6 |
Purpose: Extremely fast convergence on illustraton data for a downstream DiT model
|
|
|
|
| 17 |
|
| 18 |
Loss is all layers of DINOv2 and DINOv3 equally weighted MSE on input vs reconstruction.
|
| 19 |
|
| 20 |
+
There are no pixel losses/LPIPS/GAN used, this is DINO space loss only.
|