simple performance test about waifu2x-ncnn-vulkan

[nihui]

hi, I bring you the news of my very rough implementation :)

I may finish refactoring and polishing my code this weekend or sometime next week.

stay tuned ~

tilesize=200 input=200x200 noise=2 scale=2

cunet	one-by-one	batch
avg time(s) / image	0.86	0.0567

uprgb	one-by-one	batch
avg time(s) / image	0.75	0.0214

tilesize=200 input=1000x1000 noise=2 scale=2

cunet	one-by-one	batch
avg time(s) / image	2.26	1.28

uprgb	one-by-one	batch
avg time(s) / image	1.41	0.44

[nihui]

yeah, I've planned to make the -i/-o option accept a folder path ;)

[nihui]

https://github.com/nihui/waifu2x-ncnn-vulkan/releases/tag/20190712
just posted a new waifu2x-ncnn-vulkan release, please let me know if you encounter problems

[akai-katto]

From my understanding, this means in the future we could do bulk-upscaling (i.e upscaling a folder)?

I'm looking into using this for my waifu2x-derived project as it is much faster, but at the moment it has to re-initialize for each image. It would be a no-brainer in adopting this as a waifu2x-fork to use for my project.