Diffusers
Anzhc commited on
Commit
ebf8d2d
·
verified ·
1 Parent(s): e1b7707

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +14 -10
README.md CHANGED
@@ -203,10 +203,11 @@ Im using small test set i have on me, separated into anime(434) and photo(500) i
203
  | Anzhc MS-LC-EQ-D-VR VAE B5 | <span style="color:Orange">*5.8117*</span> | <span style="color:Orange">*9.7627*</span> | <span style="color:Orange">*29.8545*</span> | 0.1112 | <span style="color:Orange">*0.9538*</span> | 36.5573 | <span style="color:Orange">*0.0080*</span> | 4.963894 |
204
  | Anzhc MS-LC-EQ-D-VR VAE B7 | <span style="color:Crimson">5.7046</span> | <span style="color:Crimson">9.5975</span> | <span style="color:Crimson">30.0106</span> | 0.0980 | <span style="color:Crimson">0.9553</span> | 39.4477 | <span style="color:Crimson">0.0071</span> | <span style="color:Crimson">4.017592</span> |
205
 
206
- | VAE FLUX | L1&nbsp;↓ | L2&nbsp;↓ | PSNR&nbsp;↑ | LPIPS&nbsp;↓ | MSSSIM&nbsp;↑ | KL&nbsp;↓ | rFID&nbsp;↓ |
207
- |---|---|---|---|---|---|---|---|
208
- | FLUX VAE | <span style="color:Orange">*4.147* | <span style="color:Orange">*6.294* | <span style="color:Orange">*33.389* | <span style="color:Crimson">0.021 | <span style="color:Crimson">0.987 | <span style="color:Orange">*12.146* | <span style="color:Crimson">0.565 |
209
- | MSLCEQDVR VAEFLUX | <span style="color:Crimson">3.799 | <span style="color:Crimson">6.077 | <span style="color:Crimson">33.807 | <span style="color:Orange">*0.032* | <span style="color:Orange">*0.986* | <span style="color:Crimson">10.992 | <span style="color:Orange">*1.692* |
 
210
 
211
 
212
  #### Noise in latents
@@ -226,8 +227,9 @@ Im using small test set i have on me, separated into anime(434) and photo(500) i
226
 
227
  | VAE FLUX | Noise&nbsp;↓ |
228
  |---|---|
229
- | FLUX VAE | <span style="color:Orange">*10.499* |
230
  | MS‑LC‑EQ‑D‑VR VAE FLUX | <span style="color:Crimson">7.635 |
 
231
 
232
  ---
233
  ### Results on a small benchmark of 434 Illustrations from Boorus
@@ -243,10 +245,11 @@ Im using small test set i have on me, separated into anime(434) and photo(500) i
243
  | Anzhc MS-LC-EQ-D-VR VAE B5 | <span style="color:Orange">*4.0998*</span> | <span style="color:Orange">*7.5481*</span> | <span style="color:Orange">*31.4378*</span> | 0.0569 | <span style="color:Orange">*0.9717*</span> | 39.8600 | <span style="color:Orange">*0.0070*</span> | 5.178428 |
244
  | Anzhc MS-LC-EQ-D-VR VAE B7 | <span style="color:Crimson">3.9949</span> | <span style="color:Crimson">7.3784</span> | <span style="color:Crimson">31.6544</span> | 0.0508 | <span style="color:Crimson">0.9731</span> | 42.8447 | <span style="color:Crimson">0.0063</span> | <span style="color:Crimson">4.216971</span> |
245
 
246
- | VAE FLUX | L1&nbsp;↓ | L2&nbsp;↓ | PSNR&nbsp;↑ | LPIPS&nbsp;↓ | MSSSIM&nbsp;↑ | KL&nbsp;↓ | rFID&nbsp;↓ |
247
- |---|---|---|---|---|---|---|---|
248
- | FLUX VAE | <span style="color:Orange">*3.060* | <span style="color:Crimson">4.775 | <span style="color:Crimson">35.440 | <span style="color:Crimson">0.011 | <span style="color:Crimson">0.991 | <span style="color:Orange">*12.472* | <span style="color:Crimson">0.670 |
249
- | MSLCEQDVR VAEFLUX | <span style="color:Crimson">2.933 | <span style="color:Orange">*4.856* | <span style="color:Orange">*35.251* | <span style="color:Orange">*0.018* | <span style="color:Orange">*0.990* | <span style="color:Crimson">11.225 | <span style="color:Orange">*1.561* |
 
250
 
251
 
252
  #### Noise in latents
@@ -266,8 +269,9 @@ Im using small test set i have on me, separated into anime(434) and photo(500) i
266
 
267
  | VAE FLUX | Noise&nbsp;↓ |
268
  |---|---|
269
- | FLUX VAE | <span style="color:Orange">*9.913* |
270
  | MS‑LC‑EQ‑D‑VR VAE FLUX | <span style="color:Crimson">7.723 |
 
271
 
272
  KL loss suggests that this VAE implementation is much closer to SDXL, and likely will be a better candidate for further finetune, but that is just a theory.
273
 
 
203
  | Anzhc MS-LC-EQ-D-VR VAE B5 | <span style="color:Orange">*5.8117*</span> | <span style="color:Orange">*9.7627*</span> | <span style="color:Orange">*29.8545*</span> | 0.1112 | <span style="color:Orange">*0.9538*</span> | 36.5573 | <span style="color:Orange">*0.0080*</span> | 4.963894 |
204
  | Anzhc MS-LC-EQ-D-VR VAE B7 | <span style="color:Crimson">5.7046</span> | <span style="color:Crimson">9.5975</span> | <span style="color:Crimson">30.0106</span> | 0.0980 | <span style="color:Crimson">0.9553</span> | 39.4477 | <span style="color:Crimson">0.0071</span> | <span style="color:Crimson">4.017592</span> |
205
 
206
+ | VAE FLUX | L1&nbsp;↓ | L2&nbsp;↓ | PSNR&nbsp;↑ | LPIPS&nbsp;↓ | MS-SSIM&nbsp;↑ | KL&nbsp;↓ | CONSISTENCY&nbsp;↓ | rFID&nbsp;↓ |
207
+ |---|---|---|---|---|---|---|---|---|
208
+ | FLUX VAE | 4.1471 | 6.2940 | 33.3887 | <span style="color:Crimson">0.0209 | <span style="color:Orange">*0.9868* | 12.1461 | <span style="color:Orange">*0.0077* | <span style="color:Crimson">0.564150 |
209
+ | MS-LC-EQ-D-VR VAE FLUX | <span style="color:Orange">*3.799* | <span style="color:Orange">*6.077* | <span style="color:Orange">*33.807* | 0.032 | 0.986 | <span style="color:Crimson">10.992 | | 1.692 |
210
+ | Flux EQ v2 B1 | <span style="color:Crimson">3.4560 | <span style="color:Crimson">5.5851 | <span style="color:Crimson">34.6641 | <span style="color:Orange">*0.0281* | <span style="color:Crimson">0.9884 | <span style="color:Orange">*11.4340* | <span style="color:Crimson">0.0040 | <span style="color:Orange">*0.686061* |
211
 
212
 
213
  #### Noise in latents
 
227
 
228
  | VAE FLUX | Noise&nbsp;↓ |
229
  |---|---|
230
+ | FLUX VAE | 10.499 |
231
  | MS‑LC‑EQ‑D‑VR VAE FLUX | <span style="color:Crimson">7.635 |
232
+ | Flux EQ v2 B1 | <span style="color:Orange">*8.5019* |
233
 
234
  ---
235
  ### Results on a small benchmark of 434 Illustrations from Boorus
 
245
  | Anzhc MS-LC-EQ-D-VR VAE B5 | <span style="color:Orange">*4.0998*</span> | <span style="color:Orange">*7.5481*</span> | <span style="color:Orange">*31.4378*</span> | 0.0569 | <span style="color:Orange">*0.9717*</span> | 39.8600 | <span style="color:Orange">*0.0070*</span> | 5.178428 |
246
  | Anzhc MS-LC-EQ-D-VR VAE B7 | <span style="color:Crimson">3.9949</span> | <span style="color:Crimson">7.3784</span> | <span style="color:Crimson">31.6544</span> | 0.0508 | <span style="color:Crimson">0.9731</span> | 42.8447 | <span style="color:Crimson">0.0063</span> | <span style="color:Crimson">4.216971</span> |
247
 
248
+ | VAE FLUX | L1&nbsp;↓ | L2&nbsp;↓ | PSNR&nbsp;↑ | LPIPS&nbsp;↓ | MS-SSIM&nbsp;↑ | KL&nbsp;↓ | CONSISTENCY&nbsp;↓ | rFID&nbsp;↓ |
249
+ |---|---|---|---|---|---|---|---|---|
250
+ | FLUX VAE | 3.0600 | <span style="color:Orange">*4.7752* | <span style="color:Orange">*35.4400* | <span style="color:Crimson">0.0112 | <span style="color:Orange">*0.9905* | 12.4717 | <span style="color:Orange">*0.0079* | <span style="color:Crimson">0.669906 |
251
+ | MS-LC-EQ-D-VR VAE FLUX | <span style="color:Orange">*2.933* | 4.856 | 35.251 | 0.018 | 0.990 | <span style="color:Crimson">11.225 | | 1.561 |
252
+ | Flux EQ v2 B1 | <span style="color:Crimson">2.4825 | <span style="color:Crimson">4.2776 | <span style="color:Crimson">36.6027 | <span style="color:Orange">*0.0132* | <span style="color:Crimson">0.9916 | <span style="color:Orange">*11.6388* | <span style="color:Crimson">0.0039 | <span style="color:Orange">*0.744904* |
253
 
254
 
255
  #### Noise in latents
 
269
 
270
  | VAE FLUX | Noise&nbsp;↓ |
271
  |---|---|
272
+ | FLUX VAE | 9.913 |
273
  | MS‑LC‑EQ‑D‑VR VAE FLUX | <span style="color:Crimson">7.723 |
274
+ | Flux EQ v2 B1 | <span style="color:Orange">*8.4004* |
275
 
276
  KL loss suggests that this VAE implementation is much closer to SDXL, and likely will be a better candidate for further finetune, but that is just a theory.
277