Closed FrankyBoy closed 8 years ago
There is an english repo, not sure if it's the exact same tool, but you do get the cuda-powered waifu2x algorithm: https://github.com/nagadomi/waifu2x. I'd gladly help translating this repo, but my japanese is still too weak.
メッセージを英訳してくださった方がいるので、余裕ができたらやろうと思っています。 https://github.com/paul70078/waifu2x-caffe/commit/1ddc7d3903b3483e6d9366bb659651a22847a96b
@FrankyBoy Just found this
↑That is the illustration I posted on videohelp.com for an early version of this software: http://forum.videohelp.com/threads/372157-New-upscaling-algorithm-waifu2x
Translating the GUI directly is possible, but will be a maintenance nightmare unless all UI strings and Error message strings are polled into a centralized place (e.g. String Resource, xml...) in this repo first.
Think about what to do when a new version comes out with a slight UI change, before you work on it.
@MaverickTse Thanks for clarification. The new version has some additional options. Could you please provide a new translation picture for that?
Here you are (for the new features only)
@MaverickTse I really appreciate it. Thanks!
On the topic, the CLI version works wonders too, setup a batch file once and you're good. I put a few shortcuts with preset scales on the shell:sendto menu, then it's a matter of Right click > send to > batch processing.
With the added bonus of in case of crash, you can select only the missing ones after since it won't stop them all.
I can put my scripts on a gist if there's interest.
Quickly assembled up an easy-to-localize GUI for this: https://github.com/MaverickTse/waifu2x_caffe_multilang_gui/releases
see if it crash on your machine :-)
@MaverickTse In the newer versions of waifu2x-caffe there is a parameter added called TTA or test time augmentation (but I need a clear and simple definition of TTA). I tested it, it seems TTA makes the image more smoother than without using TTA however it takes longer time and I just tested a 900x1200 image and I tried to upscale x9 with TTA however waifu2x-caffe crashed (ver.1.0.8.1 using cuDNN). Without TTA it can upscale the image (900x1200) to 8100*10800
Translated from the readme: TTA: Increase PSNR by around 0.15 but 8 times slower
Update I was finally able to upscale with TTA without crashing by lowering the block size to 64 (before it was in the hundreds) but oh boy waifu2x-caffe reached a peak working set memory of 13.9 Million K with a peak commit size of 30 Million K. It took a full 37 minutes to complete it. I noticed after that my C drive has consumed space even though waifu2x-caffe is located in another drive what could be the associated file that consumed the space on my C drive? (It consumed like more than 5GB of space)
GUI now supports English ver 1.1.0. 何か翻訳に問題があったらIssuesでお知らせください。
because other people might want to work on it, too.