Correct latent scalig factors.
#3
by
Disty0
- opened
Current scaling factor of AuraVAE (0.18215
) is very wrong. This PR corrects them with adding mean and std values per channel and setting the average scaling factor to the correct average.
AuraVAE's average std is very close to 1 and average mean is very close to 0, so both average scaling factor and average shifting factor can be safely ignored (aka set them to 1 and 0).
It is better to use per channel scaling / shifting with AuraVAE because of channel 3 being an outlier compared to the rest of the channels.
Here are the statistics computed over Danbooru id 8m to 8.5m resized to around 256x256 while keeping the aspect ratio.
------------------------------
Latents Mean: [0.2539, 0.1431, 0.1484, -0.3048, -0.0985, -0.162, 0.1403, 0.2034, -0.1419, 0.2646, 0.0655, 0.0061, 0.1555, 0.0506, 0.0129, -0.1948]
Latents STD: [0.8123, 0.7376, 0.7354, 1.1827, 0.8387, 0.8735, 0.8705, 0.8142, 0.8076, 0.7409, 0.7655, 0.8731, 0.8087, 0.7058, 0.8087, 0.7615]
------------------------------
------------------------------
Channel: All
------------------------------
Avg Mean: 0.03389404746175166
Avg Max: 3.964326967376758
Avg Min: -3.8520306012579253
Avg STD: 0.8897262148301661
------------------------------
Max Mean: 0.5204
Min Mean: -0.3993
Max STD: 1.2424
Min STD: 0.6678
------------------------------
Max Max: 9.1328
Min Max: 2.3926
Max Min: -2.0527
Min Min: -7.9727
------------------------------
Scale: 1.1239
Shift: 0.0339
------------------------------
------------------------------
Channel: 0
------------------------------
Avg Mean: 0.2539401613654924
Avg Max: 3.069156550032497
Avg Min: -2.37588465545347
Avg STD: 0.8123111023897229
------------------------------
Max Mean: 1.6433
Min Mean: -1.6459
Max STD: 1.7695
Min STD: 0.3501
------------------------------
Max Max: 6.6523
Min Max: 1.291
Max Min: -0.0673
Min Min: -7.0781
------------------------------
Scale: 1.2311
Shift: 0.2539
------------------------------
------------------------------
Channel: 1
------------------------------
Avg Mean: 0.1431159930563159
Avg Max: 2.802934907645524
Avg Min: -2.2796983559316106
Avg STD: 0.737622362641964
------------------------------
Max Mean: 1.3124
Min Mean: -1.2213
Max STD: 1.2286
Min STD: 0.2706
------------------------------
Max Max: 6.0078
Min Max: 0.3081
Max Min: -0.5156
Min Min: -5.5312
------------------------------
Scale: 1.3557
Shift: 0.1431
------------------------------
------------------------------
Channel: 2
------------------------------
Avg Mean: 0.14837673422297976
Avg Max: 2.8625147102479653
Avg Min: -2.324044867627479
Avg STD: 0.7353840310769202
------------------------------
Max Mean: 0.8343
Min Mean: -0.7366
Max STD: 1.1133
Min STD: 0.3186
------------------------------
Max Max: 6.0312
Min Max: 1.1641
Max Min: -0.8223
Min Min: -5.1914
------------------------------
Scale: 1.3598
Shift: 0.1484
------------------------------
------------------------------
Channel: 3
------------------------------
Avg Mean: -0.30481267334976886
Avg Max: 3.1169687879087893
Avg Min: -3.6120808255423738
Avg STD: 1.1826804067115022
------------------------------
Max Mean: 2.9732
Min Mean: -2.3869
Max STD: 2.6592
Min STD: 0.2017
------------------------------
Max Max: 7.9961
Min Max: -0.4897
Max Min: 1.7295
Min Min: -7.8125
------------------------------
Scale: 0.8455
Shift: -0.3048
------------------------------
------------------------------
Channel: 4
------------------------------
Avg Mean: -0.09849032618338674
Avg Max: 2.3996397847847444
Avg Min: -2.8236285800364325
Avg STD: 0.8387453809086128
------------------------------
Max Mean: 1.3772
Min Mean: -1.5184
Max STD: 1.3289
Min STD: 0.2863
------------------------------
Max Max: 5.0117
Min Max: 0.3479
Max Min: -0.6899
Min Min: -5.5898
------------------------------
Scale: 1.1923
Shift: -0.0985
------------------------------
------------------------------
Channel: 5
------------------------------
Avg Mean: -0.1620394673770191
Avg Max: 2.6114278990886417
Avg Min: -3.083444767764478
Avg STD: 0.8735478430986665
------------------------------
Max Mean: 1.9843
Min Mean: -1.6738
Max STD: 1.6909
Min STD: 0.2718
------------------------------
Max Max: 6.4102
Min Max: 0.3201
Max Min: 0.0723
Min Min: -6.9648
------------------------------
Scale: 1.1448
Shift: -0.162
------------------------------
------------------------------
Channel: 6
------------------------------
Avg Mean: 0.14031346795902205
Avg Max: 3.323834901655158
Avg Min: -2.343676663007887
Avg STD: 0.8704605017097239
------------------------------
Max Mean: 1.0753
Min Mean: -1.2049
Max STD: 1.3644
Min STD: 0.3912
------------------------------
Max Max: 6.4336
Min Max: 1.7334
Max Min: -0.2607
Min Min: -4.8008
------------------------------
Scale: 1.1488
Shift: 0.1403
------------------------------
------------------------------
Channel: 7
------------------------------
Avg Mean: 0.20342172861401583
Avg Max: 3.1569343693267204
Avg Min: -2.679967695301359
Avg STD: 0.8141539146998797
------------------------------
Max Mean: 1.5621
Min Mean: -1.3831
Max STD: 1.3411
Min STD: 0.3343
------------------------------
Max Max: 6.8711
Min Max: 0.9736
Max Min: -0.9307
Min Min: -5.5781
------------------------------
Scale: 1.2283
Shift: 0.2034
------------------------------
------------------------------
Channel: 8
------------------------------
Avg Mean: -0.14187879245560003
Avg Max: 2.560529442065319
Avg Min: -2.6306781822611827
Avg STD: 0.8076236662991515
------------------------------
Max Mean: 3.1456
Min Mean: -2.6637
Max STD: 2.5096
Min STD: 0.1858
------------------------------
Max Max: 9.1328
Min Max: -0.835
Max Min: 0.5249
Min Min: -7.9727
------------------------------
Scale: 1.2382
Shift: -0.1419
------------------------------
------------------------------
Channel: 9
------------------------------
Avg Mean: 0.26459734012718117
Avg Max: 3.0655110320003813
Avg Min: -2.3551008411493757
Avg STD: 0.7409148014774868
------------------------------
Max Mean: 2.1581
Min Mean: -1.5272
Max STD: 1.6334
Min STD: 0.2518
------------------------------
Max Max: 5.9766
Min Max: 0.5698
Max Min: -0.2952
Min Min: -6.1641
------------------------------
Scale: 1.3497
Shift: 0.2646
------------------------------
------------------------------
Channel: 10
------------------------------
Avg Mean: 0.06545502249415208
Avg Max: 2.8959377155331993
Avg Min: -2.6332786419625913
Avg STD: 0.7655383823541327
------------------------------
Max Mean: 1.0873
Min Mean: -0.9293
Max STD: 1.0886
Min STD: 0.255
------------------------------
Max Max: 5.8984
Min Max: 1.1533
Max Min: -0.9204
Min Min: -5.6367
------------------------------
Scale: 1.3063
Shift: 0.0655
------------------------------
------------------------------
Channel: 11
------------------------------
Avg Mean: 0.006075071204008039
Avg Max: 3.104691862083434
Avg Min: -2.5514190991852073
Avg STD: 0.8730571919837967
------------------------------
Max Mean: 1.4358
Min Mean: -1.3148
Max STD: 1.3039
Min STD: 0.4201
------------------------------
Max Max: 6.0742
Min Max: 1.3604
Max Min: 0.0266
Min Min: -5.1992
------------------------------
Scale: 1.1454
Shift: 0.0061
------------------------------
------------------------------
Channel: 12
------------------------------
Avg Mean: 0.15552403550942404
Avg Max: 3.3750333080206802
Avg Min: -2.453994921772161
Avg STD: 0.8086856054322407
------------------------------
Max Mean: 1.3657
Min Mean: -1.423
Max STD: 1.257
Min STD: 0.3186
------------------------------
Max Max: 7.7266
Min Max: 0.5767
Max Min: -0.541
Min Min: -5.2656
------------------------------
Scale: 1.2366
Shift: 0.1555
------------------------------
------------------------------
Channel: 13
------------------------------
Avg Mean: 0.05062178046275393
Avg Max: 2.5563079486501366
Avg Min: -2.278517912629954
Avg STD: 0.705840925701482
------------------------------
Max Mean: 1.8907
Min Mean: -1.6216
Max STD: 1.628
Min STD: 0.2938
------------------------------
Max Max: 5.9766
Min Max: 0.6035
Max Min: -0.3389
Min Min: -5.4883
------------------------------
Scale: 1.4167
Shift: 0.0506
------------------------------
------------------------------
Channel: 14
------------------------------
Avg Mean: 0.012909470293981318
Avg Max: 2.5710583648477434
Avg Min: -2.8821286191580646
Avg STD: 0.8086932310000542
------------------------------
Max Mean: 1.2211
Min Mean: -1.4804
Max STD: 1.301
Min STD: 0.3582
------------------------------
Max Max: 5.9453
Min Max: 0.6626
Max Min: -0.9707
Min Min: -5.832
------------------------------
Scale: 1.2366
Shift: 0.0129
------------------------------
------------------------------
Channel: 15
------------------------------
Avg Mean: -0.1948247864819424
Avg Max: 2.2669437439496445
Avg Min: -2.944399481229158
Avg STD: 0.7615035314306112
------------------------------
Max Mean: 0.6272
Min Mean: -0.7218
Max STD: 1.0702
Min STD: 0.3168
------------------------------
Max Max: 4.9922
Min Max: 0.9253
Max Min: -1.2979
Min Min: -6.457
------------------------------
Scale: 1.3132
Shift: -0.1948
------------------------------