wl-amigo / waifu2x-converter-cpp Goto Github PK

waifu2x(original : https://github.com/nagadomi/waifu2x) re-implementation in C++ using OpenCV [NO LONGER UPDATED]

License: Other

C++ 89.04% Python 2.58% Lua 8.38%

waifu2x-converter-cpp's Introduction

waifu2x (converter only version)

This is a reimplementation of waifu2x (original) converter function, in C++, using OpenCV. This is also a reimplementation of waifu2x python version by Hector Martin. You can use this as command-line tool of image noise reduction or/and scaling.

Prebuilt binary-form release

Please see releases of this repository. There is only for-windows binary, now. Sorry.

works using waifu2x-converter

waifu2x_win_koroshell
- waifu2x-converter GUI frontend that is easy to use, and so cute. You need only drag&drop to convert your image. (and also you can set converting mode, noise reduction level, scale ratio, etc..., on GUI)
- Both waifu2x-converter x86 and x64 are included this package, and GUI see your windows architecture(x86|x64) and selects automatically which to use.
- For windows only.

Dependencies

Platform

Ubuntu
Mac OS X?
Windows

(This program probably can be built under MacOSX, because OpenCV and other libraries support OS X)

Libraries

OpenCV(C++, version 3.0.0 rc1)

This programs also depends on libraries shown below, but these are already included in this repository. CUDA Support in OpenCV is optional, since not required. (in version 1.0.0, CUDA Support is not used.)

How to build

for Ubuntu

Sorry, under construction...

These are hints for building :

I recommend to install OpenCV from sources. (build instruction is found here)
include path : include/ (/path/to/opencv/installed/directory)/include
library path : (/path/to/opencv/installed/directory)/lib
- if you have built and installed OpenCV from source, and have changed install directory(by using CMAKE_INSTALL_PREFIX), you may need to set environment variable LD_LIBRARY_PATH for your OpenCV installed directory.
libraries to link : opencv_core opencv_imgproc opencv_imgcodecs opencv_features2d
standard of C++ : c++11

Usage

Usage of this program can be seen by executing this with --help option.

(My native language is not English, then I'm sorry for my broken English.)

waifu2x-converter-cpp's People

Contributors

Stargazers

Watchers

waifu2x-converter-cpp's Issues

How to use GPU-acceleration?

I'm using pre-built binaries on Windows. What should I do to make it faster? I hoped it would be faster than original implementation in LUA, but it's awfully slow.

error LNK2019: unresolved external symbol "int cdecl cv::hal::normHamming(unsigned char const *,int)" (?normHamming@hal@cv@@YAHPBEH@Z)，该符号在函数 "double cdecl cv::norm(class cv::_InputArray const &,int,class cv::_InputArray const &)" (?norm@cv@@YANABV_InputArray@1@H0@Z)

as the title mentioned, I can't successfully compile this code

How do I convert multiple images?

I wonder how do I convert multiple images? (My intention is to convert a video to images in sequence, and then do the upscaling of the images)

Is there any method to improve this algorithm

as mentioned above, what do you think is the drawback of this algorithm? how to improve it?

How to build it properly on Linux

Anyone experiencing issues with this while building on Linux?

What kinda modification is required in this case?

Best

Add an Art Model/Scans

Could you write a MakeFile (or configure)?

I have no idea how to compile this, since I don't know cpp. I think there are many Linux user who don't know cpp either can't use this.

Since this is no longer updated, I forked and updated it.

I forked a fork of a fork of this.

See: https://github.com/DeadSix27/waifu2x-converter-cpp/

Works fine on AMD GPUs and APUs on Ubuntu 16.10 and Windows 64
I updated the OpenCV library support to 3.2 (newest at point of writing)
Added a proper build guide for windows and Ubuntu (16.10) and fixed some binary file handling.

Code also is now compiling fine on VS2015 (incl. Community version)
Pre-built binaries are also offered.

Support is appreciated, I will try doing my best to keep this at least working, (e.g for the upcoming VS2017 as well)
Open an Issue on my fork if you have any questions.

PS: If you have an nVidia GPU and would like to help me getting it to work on that as well, open an issue on my fork.

Thanks for your repository!

Hello @WL-Amigo,

I would like to re-implement waifu2x to Objective-C. May I use some of your code?

The other thing, it has very very bad performance on iOS, haha.
Maybe I will try to tune the performance more better. For the OTAKU world!
thank you.

Alpha channel support?

Could it process the PNG's alpha channel and process it separately from the color channels?

Related issue: nagadomi/waifu2x#27

hello, I am compiling your program, but I have a preblem.

There are some strange lines to me such as

static const char prog[] =
#include "modelHandler_OpenCL.cl.h"
;

static const char prog20[] =
#include "modelHandler_CUDA.ptx20.h"
;
static const char prog30[] =
#include "modelHandler_CUDA.ptx30.h"
;

When I compile, I got error because of the lines.
What happen?
It is wrong on my C++ background.
How did you use well?

RGB⇔YUV変換
各所でRGB⇔YUV変換していますが、毎回変換するのはロスがあるので、最初と最後にされたほうがよいのではないでしょうか。
UV成分のbicubic変換
Yについては2xする度に処理するので必要ですが、UV成分については別途一発で期待解像度に持っていったほうが効率がよいのではないでしょうか。

一応、Avisynth版のほうではそう考えて実装しました。私の認識が間違っていなければですが、参考になるかもしれません。