recoilme commited on
Commit
228c406
·
verified ·
1 Parent(s): 44fac4d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +16 -4
README.md CHANGED
@@ -1,7 +1,7 @@
1
  ---
2
  license: apache-2.0
3
  ---
4
- === Eval ===
5
  ```
6
  SD15 VAE | MSE=2.732e-03 PSNR=28.10 LPIPS=0.147 Edge=0.206 KL=19.821 | Z[min/mean/max/std]=[-17.375, 0.072, 16.203, 0.900] | Skew[min/mean/max]=[-0.543, -0.126, 0.070] | Kurt[min/mean/max]=[-0.151, 1.228, 4.574]
7
  SDXL VAE fp16 fix | MSE=2.018e-03 PSNR=29.67 LPIPS=0.124 Edge=0.188 KL=32.222 | Z[min/mean/max/std]=[-4.066, -0.014, 4.301, 0.861] | Skew[min/mean/max]=[-0.017, 0.105, 0.165] | Kurt[min/mean/max]=[-0.380, -0.228, -0.107]
@@ -15,7 +15,7 @@ AuraDiffusion/16ch-vae | MSE=5.361e-04 PSNR=35.80 LPIPS=0.041 Edge=0.100 KL=
15
  FLUX.1-schnell VAE | MSE=4.594e-04 PSNR=35.87 LPIPS=0.035 Edge=0.088 KL=13.016 | Z[min/mean/max/std]=[-5.824, -0.076, 6.246, 0.945] | Skew[min/mean/max]=[-0.268, 0.048, 0.483] | Kurt[min/mean/max]=[-0.498, 0.037, 0.568]
16
  AiArtLab/simplevae | MSE=4.818e-04 PSNR=36.20 LPIPS=0.035 Edge=0.095 KL=4.032 | Z[min/mean/max/std]=[-7.762, -0.061, 9.914, 0.965] | Skew[min/mean/max]=[-0.320, 0.044, 0.411] | Kurt[min/mean/max]=[-0.045, 0.346, 0.696]
17
  ```
18
- === Percent ===
19
  ```
20
  | Model | PSNR | LPIPS | Edge |
21
  |----------------------------|-----------|-----------|-----------|
@@ -32,6 +32,18 @@ AiArtLab/simplevae | MSE=4.818e-04 PSNR=36.20 LPIPS=0.035 Edge=0.095 KL=
32
  | AiArtLab/simplevae | 128.8% | 415.2% | 217.7% |
33
  ```
34
 
35
- Compare
36
 
37
- https://imgsli.com/NDE1MzE0/5/2
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
  ---
4
+ ## === Eval ===
5
  ```
6
  SD15 VAE | MSE=2.732e-03 PSNR=28.10 LPIPS=0.147 Edge=0.206 KL=19.821 | Z[min/mean/max/std]=[-17.375, 0.072, 16.203, 0.900] | Skew[min/mean/max]=[-0.543, -0.126, 0.070] | Kurt[min/mean/max]=[-0.151, 1.228, 4.574]
7
  SDXL VAE fp16 fix | MSE=2.018e-03 PSNR=29.67 LPIPS=0.124 Edge=0.188 KL=32.222 | Z[min/mean/max/std]=[-4.066, -0.014, 4.301, 0.861] | Skew[min/mean/max]=[-0.017, 0.105, 0.165] | Kurt[min/mean/max]=[-0.380, -0.228, -0.107]
 
15
  FLUX.1-schnell VAE | MSE=4.594e-04 PSNR=35.87 LPIPS=0.035 Edge=0.088 KL=13.016 | Z[min/mean/max/std]=[-5.824, -0.076, 6.246, 0.945] | Skew[min/mean/max]=[-0.268, 0.048, 0.483] | Kurt[min/mean/max]=[-0.498, 0.037, 0.568]
16
  AiArtLab/simplevae | MSE=4.818e-04 PSNR=36.20 LPIPS=0.035 Edge=0.095 KL=4.032 | Z[min/mean/max/std]=[-7.762, -0.061, 9.914, 0.965] | Skew[min/mean/max]=[-0.320, 0.044, 0.411] | Kurt[min/mean/max]=[-0.045, 0.346, 0.696]
17
  ```
18
+ ## === Percent ===
19
  ```
20
  | Model | PSNR | LPIPS | Edge |
21
  |----------------------------|-----------|-----------|-----------|
 
32
  | AiArtLab/simplevae | 128.8% | 415.2% | 217.7% |
33
  ```
34
 
35
+ ## Compare
36
 
37
+ https://imgsli.com/NDE1MzE0/5/2
38
+
39
+ ## VAE Training Process
40
+
41
+ - Inited from AuraDiffusion/16ch-vae (not compatible), added mid block/retrained
42
+ - Dataset: 100,000 PNG images
43
+ - Training Time: ~ 2 weeks
44
+ - Hardware: Single RTX 5090
45
+ - Resolution: 512px
46
+ - Precision: FP32
47
+ - Effective Batch Size: 16
48
+ - Optimizer: AdamW (8-bit)
49
+ - Balanced losses (lpips, MSE, MAE, Edge, KL)