This repository contains code for the paper Escaping Plato’s Cave: 3D Shape from Adversarial Rendering (ICCV2019).
More detailed information and results can be found on our project page.
For training you can use any image collection you would like. For now two example datasets "tree" and "chanterelle" are provided under 'datasets/.
If you have your own data set go to 'scripts/data' and have a look at the already implemented custom implementations for data sets.
You may want to adjust them accordingly or create new ones.
In case you want to create a new Dataset class you should do so under scripts/data.
- Linux (not tested for MacOS or Windows)
- Python3
- CPU or NVIDIA GPU
Clone this repo:
git clone https://github.com/henzler/platonicgan
cd platonicgan
In order to install the required python packages run (in a new virtual environment):
pip install -r requirements.txt
Note: The model was trained with PyTorch 1.0.0, but also tested with Pytorch 1.3.1
For training execute the following command (for default config, currently this will train the tree dataset):
python train.py
or for a specific config:
python train.py --config_file=scripts/configs/your_config_file.yaml
Use tensorboard to visualise intermediate training results:
tensorboard --logdir=.
If you use this code for your research, please cite our paper.
@InProceedings{henzler2019platonicgan,
author = {Henzler, Philipp and Mitra, Niloy J. and Ritschel, Tobias},
title = {Escaping Plato's Cave: 3D Shape From Adversarial Rendering},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}