Public Repository

Last pushed: 3 years ago
Short Description
A working waifu2x web server
Full Description


This image contains a ubuntu 14.04 base image with CUDA 7.0.28, NVIDIA 352.21, Torch 7 and everything needed to run a waifu2x web server.
This image has been tested and works on Arch Linux host (CUDA 7.0.28 and NVIDIA 352.21 installed).

What is waifu2x? waifu2x is a super upscaling and denoising algorithm, that's all I know. waifu2x repo

Why use this docker image instead using

  • The demo application has a strict restriction (Size: 2MB, Noise Reduction: 2560x2560px, Upscaling: 1280x1280px) to the image uploaded because of their resource limitation.
  • Not everybody likes Amazon / owns EC2 account / knows what public AMI is / how to use them. (just like me ;_;)
  • The waifu2x web server's limitation of this docker image has already been adjusted (Size: 100MB, Noise Reduction: 20000x20000px, Upscaling: 10000x10000px). By hosting your own server, feel free to upload a 5000x5000 image and go waifu2x.

Requirements to run this image

  • Host: Linux distro with CUDA 7.0.28 and NVIDIA 352.21 (deviceQuery will show FAIL if any one of this mismatch)
  • Graphic card: CUDA compatible NVIDIA graphic card (Or you may figure out your own way to emulate a CUDA card)

Startup script


${CUDA_DEVICE_QUERY_PATH}/deviceQuery > /dev/null # This is to activate /dev/nvidia-uvm

docker run --device ${DEV1}:${DEV1} --device ${DEV2}:${DEV2} --device ${DEV3}:${DEV3} \
-p ${PORT}:8812 -ti saren/ubuntu-waifu2x /bin/bash -c "export LD_LIBRARY_PATH='/usr/local/lib:/usr/local/cuda-7.0/lib64'; cd /root/waifu2x; th web.lua"

Running the waifu2x web server

  • Starting web server: $ ./
  • Open a web browser, go to http://localhost:8812 (or as defined in the startup script)
Docker Pull Command