yui0 / waifu2x-ocl

Fast waifu2x converter with GPU optimization

waifu2x-ocl

Fast waifu2x converter with GPU optimization. Using OpenCL.

Platform

• Linux with OpenCL
• macOS with OpenCL
• Windows with OpenCL

Download

• You can download this program from here.

How to build on macOS

$ make

How to build on Linux

$ make

How to build on Windows

1. Installing OpenCL

• nVidia: https://developer.nvidia.com/cuda-downloads
• AMD: http://developer.amd.com/tools/hc/AMDAPPSDK/downloads/Pages/default.aspx
• Intel: Intel SDK for OpenCL

2. Including OpenCL in the Project

• In the ‘Additional Include Directories’ field in ‘VC++ Directories’ add the following information depending on your graphics card.
• nVidia: - “$(CUDA_INC_PATH)”
• AMD: - “$(AMDAPPSDKROOT)\include”
• Intel: - “C:\Intel\OpenCL\sdk\include”

3. Linking OpenCL

• Choose ‘Linker-> General’.
• In the ‘Additional Dependencies’ field enter the following, again this is based on your card vendor.
  • nVidia: - “$(CUDA_LIB_PATH)”
  • AMD: - “$(AMDAPPSDKROOT)\lib\x86” (for 64 bit users you may need to change the x86 to x86_64)
  • Intel: - “C:\Intel\OpenCL\sdk\lib\x86” (for 64 bit users you may need to change the x86 to x64)
• Still in the ‘Linker’ submenu, select ‘Input’.
• In the ‘Additional Dependencies’ field click on the arrow that appears at the end of the field and choose Edit…
• In the dialog that appears enter “OpenCL.lib”

How to use

$ ./waifu2x_ocl -h
Usage: ./waifu2x_ocl [options] file

Options:
-h                 Print this message
-m <model name>    waifu2x model name [noise2_model.json...]
-s <scale>         Magnification [1.0, 1.6, 2.0...]
-o <output name>   output file name [*.png, *.jpg]

$ ./waifu2x_ocl -s 1.0 nyanko.jpg
$ ./waifu2x_ocl -m vgg_7/art_y/noise3_model.json nyanko.jpg
$ ./waifu2x_ocl 001_200x.jpg
001_200x.jpg 200x199 3
wsize:287136
AMD Radeon HD 7800 Series (TAHITI / DRM 3.23.0 / 4.16.11-berry, LLVM 6.0.0) (platform 0, device 0)
Maximum memory allocation size is 2254046412 bytes
200 199 -> 416 414 *2.000000
0 0
159 0
0 157
159 157

Demo

Original

Normal Resize by Gimp

Waifu2x Resize

• ./waifu2x_ocl -s 2.0 -m noise3_model.json waifu_s.jpg -o waifu_d.png

References

• https://github.com/yui0/catseye
• Image Super-Resolution Using Deep Convolutional Networks
• EnhanceNet: Single Image Super-Resolution Through Automated Texture Synthesis
• Waifu2x
  • Waifu2x webgl version / GLSL code
  • Original implementation
  • https://github.com/ueshita/waifu2x-converter-glsl
  • https://stanko.github.io/super-resolution-image-resizer
  • https://www.slideshare.net/KosukeNakago/seranet
• OpenCL
  • https://developer.amd.com/wordpress/media/2012/10/Optimizations-ImageConvolution1.pdf
• Picture
  • Nyanko: https://www.illust-box.jp/member/view/7263/
  • Nyanko: http://www.poipoi.com/yakko/cgi-bin/sb/log/eid5173.html#more-5173

waifu2x-ocl's Issues

Outputs a black image

When running the program, the output is a black image with the exact color of (1, 1, 0).

segmentation fault (core dumped)

$ ./waifu2x_ocl -s 2.0 waifu_s.jpg
waifu_s.jpg 256x384 3
wsize:287136
[1]    17099 segmentation fault (core dumped)  ./waifu2x_ocl -s 2.0 waifu_s.jpg

System: Linux Mint 19.2 Tina x86_64
No GPU, using OpenCL
Built with make

Any tips how to fix it?

Assertion `!r` failed

I compiled and ran this:

./waifu2x-ocl/waifu2x_ocl -s 2.0 frames/frm0000017000.png -o frm17000.png

and all I can get is the following:

frm0000001722.png 1920x1072 3
waifu2x: waifu2x_ocl.c:462: int waifu2x_ocl(char *, char *, char *, float): Assertion `!r' failed.
Aborted (core dumped)

¯\_(ツ)_/¯

Not using iGPU on macOS 10.14 Mojave

The program uses CPU instead of integrated GPU on my mac.
Compiled master directly with make

Yakumo-Yukari:waifu2x-ocl gnattu$ ./waifu2x_ocl -s 2.0 /Users/gnattu/Pictures/66176711_p1_master1200.jpg 
/Users/gnattu/Pictures/66176711_p1_master1200.jpg 1200x675 3
wsize:287136
Intel(R) Core(TM) i7-7567U CPU @ 3.50GHz (platform 0, device 0)
Maximum memory allocation size is 4294967296 bytes