Let there be color!: joint end-to-end learning of global and local image priors for automatic image colorization with simultaneous classification

SIGGRAPH 2016


Reviews

Information

  • Paper topic: Images
  • Software type: Code
  • Able to run a replicability test: True
  • Replicability score: 5
  • Software language: Other
  • License: CC BY-NC-SA 4.0
  • Build mechanism: Not applicable (python, Matlab..)
  • Dependencies: torch / torch-nn
  • Documentation score {0,1,2}: 1
  • Reviewer: Nicolas Bonneel <nicolas.bonneel@liris.cnrs.fr>
  • Time spent for the test (build->first run, timeout at 100min): 30min

Source code information

Comments

The short LUA code comes with a downloadable pretrained network (the data to retrain the network is from http://places.csail.mit.edu/ but the code to train the network is not provided). I used the Ubuntu for Windows framework (with Windows 10) to run the code, which was simpler here.
I still needed to perform a number of steps:
- dos2unix download_model.sh to prevent including the \r in the download URL
- sudo apt-get install cmake and sudo apt-get install libreadline-dev for dependencies
- installing torch with nn was slightly harder than the instructions on the torch website claim:
git clone https://github.com/torch/distro.git ~/torch --recursive
cd ~/torch
then on line 178 of install--deps  you need to change
sudo apt-get install -y python-software-properties
to sudo apt-get install -y software-properties-common since python-software-properties is not supported anymore.
Then you can run bash install-deps and ./install.sh
But then you still need to install nn:
sudo apt install luarocks
sudo ~/torch/install/bin/luarocks install torch
sudo ~/torch/install/bin/luarocks install nn
and finally run the colorize script, not directly with "th" but using:
~/torch/install/bin/th colorize.lua ansel_colorado_1941.png out.png
Then everything works fine, runs smoothly (counts 2-3 minutes on the CPU for the only example image provided, which exactly replicate the landscape result in the teaser).

If you want to contribute with another review, please follow these instructions.

Please consider to cut/paste/edit the raw JSON data attached to this paper.