Awesome
Neural Scene De-rendering
This repository contains scripts for rendering Minecraft images as those in the CVPR'17 Neural Scene De-rendering paper.
<img src="http://nsd.csail.mit.edu/images/repo.jpg" width="600">Prerequisites
Project Malmo: Please follow the instructions from Malmo
to launch the Minecraft client (0.30.0).
Please also include the path to Malmo schemas in the environment variable MALMO_XSD_PATH
.
Torch: We use Torch 7
for our implementation with these additional packages:
libMalmoLua
: offered in Project Malmo. Its path should be included in the environment variableLUA_CPATH
.
Guide
Our current release has been tested on Ubuntu 14.04.
Cloning the repository
git clone git@github.com:jiajunwu/nsd.git
cd nsd
Connecting to the client
Please update line 101 of mc-gen.lua
with the actual IP and port of your Minecraft client.
Image rendering demo (main.lua
)
The demo calls the rendering function to render an image based on the scene XML. Please find descriptions of the scene XML we used in the source file.
th main.lua
The script renders the image and saves it as demo.png
. It should look as follows.
Sometimes the image might look darker than expected, in which cases please try again.
Reference
@inproceedings{nsd,
title={{Neural Scene De-rendering}},
author={Wu, Jiajun and Tenenbaum, Joshua B and Kohli, Pushmeet},
booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2017}
}
For any questions, please contact Jiajun Wu (jiajunwu@mit.edu).