Jonathan Ho, Evan Lohn, Pieter Abbeel
Neural Information Processing Systems, 2019
https://arxiv.org/abs/1905.08500
Contains a PyTorch implementation of Flow++.
Models available here.
- Python 3.6.7
- PyTorch 1.1.0 (CUDA 10.0)
- Install Anaconda with Python 3.6.7
conda install pytorch torchvision cudatoolkit=10.0 -c pytorch
pip install tqdm
- Extract this codebase into a directory called
compression
- Build the underlying C++ library (need a compiler with OpenMP support):
cd compression/ans
mkdir build
cd build
cmake ..
make