Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc
|
| 3 |
+
library_name: diffusers
|
| 4 |
+
tags:
|
| 5 |
+
- art
|
| 6 |
+
model-index:
|
| 7 |
+
- name: 16ch-VAE
|
| 8 |
+
results:
|
| 9 |
+
- task:
|
| 10 |
+
type: encoder-loss
|
| 11 |
+
dataset:
|
| 12 |
+
name: yerevann/coco-karpathy
|
| 13 |
+
type: image
|
| 14 |
+
metrics:
|
| 15 |
+
- name: PSNR
|
| 16 |
+
type: PSNR
|
| 17 |
+
value: 31.5151
|
| 18 |
+
---
|
| 19 |
+
|
| 20 |
+
## 16ch-VAE
|
| 21 |
+
|
| 22 |
+
> Disclaimer: this VAE is not intended to be a replacement for SD3's VAE since the latent spaces are entirely different.
|
| 23 |
+
|
| 24 |
+
A fully open source 16ch VAE reproduction for the [SD3](https://arxiv.org/abs/2403.03206). Useful for people who are building their own image generation models and need an off-the-shelf VAE
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
| VAE | rFID | PSNR | LPIPS |
|
| 28 |
+
|------------------|--------|---------|--------|
|
| 29 |
+
| SD1.5 VAE | 0.3131 | 26.4332 | 0.0328 |
|
| 30 |
+
| SDXL VAE | 0.3511 | 26.7577 | 0.032 |
|
| 31 |
+
| SD3 VAE | 0.0257 | 30.3231 | 0.0132 |
|
| 32 |
+
| [16ch-VAE](https://huggingface.co/AuraDiffusion/16ch-vae) | 0.0667 | 31.5151 | 0.0136 |
|
| 33 |
+
| [16ch-VAE with FFT](https://huggingface.co/AuraDiffusion/16ch-vae)* | 0.1584 | 31.0542 | 0.0281 |
|
| 34 |
+
|
| 35 |
+
|
| 36 |
+
### Usage
|
| 37 |
+
|
| 38 |
+
Awaiting https://github.com/huggingface/diffusers/pull/8769 in diffusers!
|