neural-style-initializations.txt

# start from original image
x = img_arr.copy()

Current loss value: 0.0
Current loss value: 0.128024756908
Current loss value: 0.0810014903545
Current loss value: 0.060525547713
Current loss value: 0.0497511178255
Current loss value: 0.0411454588175
Current loss value: 0.0351021699607
Current loss value: 0.0371176600456
Current loss value: 0.0275756139308
Current loss value: 0.0271842237562 
...

# start from noise
x = rand_img(shp)

Current loss value: 58.9655227661
Current loss value: 19.7473297119
Current loss value: 10.248167038
Current loss value: 6.65550661087
Current loss value: 5.05227422714
Current loss value: 4.09817075729
Current loss value: 3.44871687889
Current loss value: 3.18568682671
Current loss value: 3.18331480026
Current loss value: 3.18332576752
...

# in between original and noise
x = 0.5 * (x + img_arr.copy())

Current loss value: 3.88481450081
Current loss value: 2.04441308975
Current loss value: 1.4591691494
Current loss value: 1.18392443657
Current loss value: 1.0077599287
Current loss value: 0.883792936802
Current loss value: 0.790984511375
Current loss value: 0.719243049622
Current loss value: 0.663124024868
Current loss value: 0.617825627327
Current loss value: 0.578325688839
Current loss value: 0.544580042362
Current loss value: 0.514340817928
Current loss value: 0.488735228777
Current loss value: 0.46531945467
Current loss value: 0.443997949362
Current loss value: 0.424483478069
Current loss value: 0.407500714064
Current loss value: 0.389991462231
Current loss value: 0.374011278152
Current loss value: 0.373741924763
Current loss value: 0.37374189496
Current loss value: 0.373741924763
Current loss value: 0.37374189496
...

Loss essentially stops changing after the ...