Awesome
Creative Birds: Self-Supervised Single-View 3D Style Transfer
Visualizations
- Evolutionary Birds from Different Angles
- Shape transfer with Different Scale Factor (Alpha)
- Texture transfer with Different Switch Gate
Prerequisites
- Download code & pre-trained model: Git clone the code by:
git clone https://github.com/anonymous-226/2D-to-3D-Evolution-Transfer $ROOTPATH
The pretrained model can be found from here, which should be unzipped in $ROOTPATH
.
- Install packages:
conda create -n evo_trans python=3.6
conda activate evo_trans
conda install pytorch==1.8.0 torchvision==0.9.0 cudatoolkit=11.1 -c pytorch -c conda-forge
pip install -r requirements.txt
- Install external tools:
cd $ROOTPATH/2D-to-3D-Evolution-Transfer/dependency
unzip meshzoo-0.4.3.zip
cd meshzoo-0.4.3
python setup.py install
cd $ROOTPATH/2D-to-3D-Evolution-Transfer/dependency
unzip neural_renderer.zip
cd neural_renderer
python setup.py install
Run the test
Run the following command from the $ROOTPATH/2D-to-3D-Evolution-Transfer
directory:
python -m evo_trans.experiments.test_df2
The result can be found in $ROOTPATH/2D-to-3D-Evolution-Transfer/evo_trans/cachedir/visualization
directory.