nihui/srmd-ncnn-vulkan

SRMD super resolution implemented with ncnn library

SRMD ncnn Vulkan

ncnn implementation of SRMD super resolution.

srmd-ncnn-vulkan uses ncnn project as the universal neural network inference framework.

Download

Download Windows/Linux/MacOS Executable for Intel/AMD/Nvidia GPU

https://github.com/nihui/srmd-ncnn-vulkan/releases

This package includes all the binaries and models required. It is portable, so no CUDA or Caffe runtime environment is needed :)

Usages

Example Command

srmd-ncnn-vulkan.exe -i input.jpg -o output.png -n 3 -s 2

Full Usages

Usage: srmd-ncnn-vulkan -i infile -o outfile [options]...

  -h                   show this help
  -v                   verbose output
  -i input-path        input image path (jpg/png/webp) or directory
  -o output-path       output image path (jpg/png/webp) or directory
  -n noise-level       denoise level (-1/0/1/2/3/4/5/6/7/8/9/10, default=3)
  -s scale             upscale ratio (2/3/4, default=2)
  -t tile-size         tile size (>=32/0=auto, default=0) can be 0,0,0 for multi-gpu
  -m model-path        srmd model path (default=models-srmd)
  -g gpu-id            gpu device to use (default=0) can be 0,1,2 for multi-gpu
  -j load:proc:save    thread count for load/proc/save (default=1:2:2) can be 1:2,2,2:2 for multi-gpu
  -x                   enable tta mode
  -f format            output image format (jpg/png/webp, default=ext/png)

• input-path and output-path accept either file path or directory path
• noise-level = noise level, large value means strong denoise effect, -1 = no effect
• scale = scale level, 2 = upscale 2x, 3 = upscale 3x, 4 = upscale 4x
• tile-size = tile size, use smaller value to reduce GPU memory usage, default selects automatically
• load:proc:save = thread count for the three stages (image decoding + waifu2x upscaling + image encoding), using larger values may increase GPU usage and consume more GPU memory. You can tune this configuration with "4:4:4" for many small-size images, and "2:2:2" for large-size images. The default setting usually works fine for most situations. If you find that your GPU is hungry, try increasing thread count to achieve faster processing.
• format = the format of the image to be output, png is better supported, however webp generally yields smaller file sizes, both are losslessly encoded

If you encounter a crash or error, try upgrading your GPU driver:

• Intel: https://downloadcenter.intel.com/product/80939/Graphics-Drivers
• AMD: https://www.amd.com/en/support
• NVIDIA: https://www.nvidia.com/Download/index.aspx

Sample Images

Original Image

Upscale 4x with ImageMagick Lanczo4 Filter

convert origin.jpg -resize 400% output.png

Upscale 4x with waifu2x scale=2 model=upconv_7_photo twice

waifu2x-ncnn-vulkan.exe -i origin.jpg -o 2x.png -s 2 -m models-upconv_7_photo
waifu2x-ncnn-vulkan.exe -i 2x.png -o 4x.png -s 2 -m models-upconv_7_photo

Upscale 4x with srmd noise=3 scale=4

srmd-ncnn-vulkan.exe -i origin.jpg -o output.png -n 3 -s 4

Original SRMD Project

• https://github.com/cszn/SRMD
• https://github.com/cszn/KAIR

Other Open-Source Code Used

• https://github.com/Tencent/ncnn for fast neural network inference on ALL PLATFORMS
• https://github.com/webmproject/libwebp for encoding and decoding Webp images on ALL PLATFORMS
• https://github.com/nothings/stb for decoding and encoding image on Linux / MacOS
• https://github.com/tronkko/dirent for listing files in directory on Windows

srmd-ncnn-vulkan's Issues

Your Results in New Super-Resolution Benchmarks

Hello,

MSU Graphics & Media Lab Video Group has recently launched two new Super-Resolution Benchmarks.

• Video Upscalers Benchmark: Quality Enhancement determines the best upscaling methods for increasing video resolution and improving visual quality.
• Super-Resolution for Video Compression benchmark aims to test Super-Resolution methods on compressed videos and select the best model for each video codec standard.

Your method achieved 19th place in Video Upscalers Benchmark: Quality Enhancement in 'Animation 4x' category. We look forward to your future work!

We would be grateful for your feedback on our work.

Compile error: missing "include <locale.h>" in main.cpp

suddenly got compiler error on LC_ALL and setlocale on Win10. Perhaps locale.h is missing?

[Unrelated request for ESRGAN]

Thanks for the work!
Can you consider implement ESRGAN in ncnn, please? お願い！
It seems like it's the State-of-the-art of super resolution and would be really really nice to have.

Anyway, thanks again!

Wrong number of arguments to ncnn API functions?

I get a lot of errors related to the ncnn API functions like the following when trying to compile:

ninja
[13/15] Building CXX object CMakeFiles/srmd-ncnn-vulkan.dir/srmd.cpp.o
FAILED: CMakeFiles/srmd-ncnn-vulkan.dir/srmd.cpp.o
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/c++   -I. -isystem /usr/local/include/ncnn -isystem /usr/local/include -Xclang -fopenmp -I/usr/local/opt/libomp/include -O3 -DNDEBUG -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.15.sdk -fPIE   -Xclang -fopenmp -I/usr/local/opt/libomp/include -MD -MT CMakeFiles/srmd-ncnn-vulkan.dir/srmd.cpp.o -MF CMakeFiles/srmd-ncnn-vulkan.dir/srmd.cpp.o.d -o CMakeFiles/srmd-ncnn-vulkan.dir/srmd.cpp.o -c ../srmd.cpp
../srmd.cpp:123:31: error: no matching member function for call to 'create'
                srmd_preproc->create(srmd_preproc_tta_int8s_spv_data, sizeof(srmd_preproc_tta_int8s_spv_data), "srmd_preproc_tta_int8s", specializations, 9, 10);
                ~~~~~~~~~~~~~~^~~~~~
/usr/local/include/ncnn/pipeline.h:44:9: note: candidate function not viable: requires 4 arguments, but 6 were provided
    int create(VkShaderModule shader_module, const std::vector<vk_specialization_type>& specializations, int binding_count, int push_constant_count);

The same thing happens when trying to compile nihui/waifu2x-ncnn-vulkan. In all cases, candidate function not viable: requires m arguments, but n were provided. I compiled and installed ncnn from https://github.com/Tencent/ncnn.

Do you have any insight into what's going on?

strange behavior

closed

Deblur request

Hi,

Is it possible for you to somehow make this work on image deblurring also ?

关于srmd 模型param文件里的一个问题

up ，srmd_x4.param 第一个conv的weight_data_size是21888 这个如果没有问题的话 那意味着输入图像的channel size是19？33128*19=21888？那这个输入的预处理是怎么做成19通道的呢？

Feature request: support for input and output pipes?

Would be nice if input and output pipes supporting YUV4MPEG pipes from/to ffmpeg could be supported.

vkEnumeratePhysicalDevices failed -3

run with ubuntu18.04 error but my windows pc run success, Ubuntu Nvidia RTX2080TI*2 Driver Version: 460.56
vkEnumeratePhysicalDevices failed -3
invalid gpu device

Is it possbile to porting to Android phone?

Thx.

-j 2:2:2 causes my single-GPU system to entirely crash

Can -j 2 (and similar things) be blocked on single GPU systems? I think some motherboards might be better about this but not mine. Mine does a complete restart without warning if I use -j 2 or -j 2:2:2.

I am not sure entirely what the issue and my motherboard/CPU is a bit dated, but I know that using -j 1:1:1 works fine.

scale=1

Hello, could there be an option for no scaling, and just denoising?

16bit

Can you add in or enable saving to 16bit/channel (48bit PNG)?
I have tried exporting images with AVISynth with 8 bit and 16bit, but it always saves back to 8bit/channel (24bit PNG).
Thanks for your time.