Hugging Face's logo

profaker
/
sd-vae-ft-mse-original

Diffusers
Safetensors
Model card Files Community
profaker commited on
Commit
8a040c8
·
verified ·
1 Parent(s): e471742

Delete README.md

Browse files
Files changed (1) hide show
  1. README.md +0 -73
README.md DELETED
@@ -1,73 +0,0 @@
1
- ---
2
- license: mit
3
- tags:
4
- - stable-diffusion
5
- - stable-diffusion-diffusers
6
- - text-to-image
7
- inference: false
8
- ---
9
- # Improved Autoencoders
10
-
11
- ## Utilizing
12
- These weights are intended to be used with the original [CompVis Stable Diffusion codebase](https://github.com/CompVis/stable-diffusion). If you are looking for the model to use with the 🧨 diffusers library, [come here](https://huggingface.co/CompVis/stabilityai/sd-vae-ft-ema).
13
-
14
- ## Decoder Finetuning
15
- We publish two kl-f8 autoencoder versions, finetuned from the original [kl-f8 autoencoder](https://github.com/CompVis/latent-diffusion#pretrained-autoencoding-models) on a 1:1 ratio of [LAION-Aesthetics](https://laion.ai/blog/laion-aesthetics/) and LAION-Humans, an unreleased subset containing only SFW images of humans. The intent was to fine-tune on the Stable Diffusion training set (the autoencoder was originally trained on OpenImages) but also enrich the dataset with images of humans to improve the reconstruction of faces.
16
- The first, _ft-EMA_, was resumed from the original checkpoint, trained for 313198 steps and uses EMA weights. It uses the same loss configuration as the original checkpoint (L1 + LPIPS).
17
- The second, _ft-MSE_, was resumed from _ft-EMA_ and uses EMA weights and was trained for another 280k steps using a different loss, with more emphasis
18
- on MSE reconstruction (MSE + 0.1 * LPIPS). It produces somewhat ``smoother'' outputs. The batch size for both versions was 192 (16 A100s, batch size 12 per GPU).
19
- To keep compatibility with existing models, only the decoder part was finetuned; the checkpoints can be used as a drop-in replacement for the existing autoencoder..
20
-
21
- _Original kl-f8 VAE vs f8-ft-EMA vs f8-ft-MSE_
22
-
23
- ## Evaluation
24
- ### COCO 2017 (256x256, val, 5000 images)
25
- | Model | train steps | rFID | PSNR | SSIM | PSIM | Link | Comments
26
- |----------|---------|------|--------------|---------------|---------------|-----------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|
27
- | | | | | | | | |
28
- | original | 246803 | 4.99 | 23.4 +/- 3.8 | 0.69 +/- 0.14 | 1.01 +/- 0.28 | https://ommer-lab.com/files/latent-diffusion/kl-f8.zip | as used in SD |
29
- | ft-EMA | 560001 | 4.42 | 23.8 +/- 3.9 | 0.69 +/- 0.13 | 0.96 +/- 0.27 | https://huggingface.co/stabilityai/sd-vae-ft-ema-original/resolve/main/vae-ft-ema-560000-ema-pruned.ckpt | slightly better overall, with EMA |
30
- | ft-MSE | 840001 | 4.70 | 24.5 +/- 3.7 | 0.71 +/- 0.13 | 0.92 +/- 0.27 | https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.ckpt | resumed with EMA from ft-EMA, emphasis on MSE (rec. loss = MSE + 0.1 * LPIPS), smoother outputs |
31
-
32
-
33
- ### LAION-Aesthetics 5+ (256x256, subset, 10000 images)
34
- | Model | train steps | rFID | PSNR | SSIM | PSIM | Link | Comments
35
- |----------|-----------|------|--------------|---------------|---------------|-----------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|
36
- | | | | | | | | |
37
- | original | 246803 | 2.61 | 26.0 +/- 4.4 | 0.81 +/- 0.12 | 0.75 +/- 0.36 | https://ommer-lab.com/files/latent-diffusion/kl-f8.zip | as used in SD |
38
- | ft-EMA | 560001 | 1.77 | 26.7 +/- 4.8 | 0.82 +/- 0.12 | 0.67 +/- 0.34 | https://huggingface.co/stabilityai/sd-vae-ft-ema-original/resolve/main/vae-ft-ema-560000-ema-pruned.ckpt | slightly better overall, with EMA |
39
- | ft-MSE | 840001 | 1.88 | 27.3 +/- 4.7 | 0.83 +/- 0.11 | 0.65 +/- 0.34 | https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.ckpt | resumed with EMA from ft-EMA, emphasis on MSE (rec. loss = MSE + 0.1 * LPIPS), smoother outputs |
40
-
41
-
42
- ### Visual
43
- _Visualization of reconstructions on 256x256 images from the COCO2017 validation dataset._
44
-
45
- <p align="center">
46
- <br>
47
- <b>
48
- 256x256: ft-EMA (left), ft-MSE (middle), original (right)</b>
49
- </p>
50
-
51
- <p align="center">
52
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00025_merged.png />
53
- </p>
54
-
55
- <p align="center">
56
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00011_merged.png />
57
- </p>
58
-
59
- <p align="center">
60
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00037_merged.png />
61
- </p>
62
-
63
- <p align="center">
64
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00043_merged.png />
65
- </p>
66
-
67
- <p align="center">
68
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00053_merged.png />
69
- </p>
70
-
71
- <p align="center">
72
- <img src=https://huggingface.co/stabilityai/stable-diffusion-decoder-finetune/resolve/main/eval/ae-decoder-tuning-reconstructions/merged/00029_merged.png />
73
- </p>