News

• 2023.7.17 - 🤣🤣🤣 Our paper "Neural Image Re-Exposure" has been rejected by ICCV 2023 🤪🤪🤪. We temporarily release this code for a better understanding of our paper, (specially for the understanding of Neural Film, Neural Shutter, and Exposure Module.) As our work remains to be improved, and we have some follow-up works in the full version, the released version is trimmed and has not been tested. It may lack components for deployment, which will be fixed and re-arranged in a month or two. By then, a new version of our code together with our revised version of paper will be released. ``

Important modules

The core module of our NIRE model are implemented in models/archs/NIRE_arch.py and module/temporalize_tsfm.py. We recommend read these codes for better understanding of our framework and method.

Environment Setup

The code requires:

• RTX2080Ti GPU (11G Memory)
• Python 3.8
• Pytorch 1.11.0
• torchvision 0.12.0
• cudatoolkit 11.3

apt install libgl1 libglib2.0-dev  # may miss this package in docker container 
conda create -n eventinr python=3.8
conda activate eventinr
conda install pytorch==1.11.0 torchvision==0.12.0 torchaudio==0.11.0 cudatoolkit=11.3 -c pytorch
pip install matplotlib opencv-python pillow tqdm pyyaml tensorboard imageio scikit-image numba einops
pip install argcomplete engineering_notation easygui numba h5py screeninfo  # for the event simulator
# pip install av  # (optional) for parsing aedat data
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11.0/index.html

Quick Start

python run_nire.py