Prerequisites
Our development environment is:
Linux (Ubuntu 16.04 or 18.04)
Python 3.6
CUDA 10.1
NOTE:
The main code should be able to support in Windows and macOS. (We did not experiment, but it should be possible to run if the environment is successfully installed.)
CUDA version should be greater than 9.0 (CUDA >=9.0) so that can install apex.
Installation
We recommend using anaconda to create a new environment.
Clone the repo
git clone https://github.com/BIGWangYuDong/UWEnhancement.git UW
Create a conda virtual environment and activate it.
conda create -n UW python=3.6 -y conda activate UW
Install PyTorch and torchvision following the official instructions.
conda install pytorch==1.5.0 torchvision==0.6.0 cudatoolkit=10.1 -c pytorch
Install build requirements and install addict.
pip install -r requirements.txt pip install addict
We support fp16, if want to use fp16 during training or testing, you can install apex.
git clone https://github.com/NVIDIA/apex cd apex pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./ # If the installation fails, you can try : python setup.py install