Learning Event Guided High Dynamic Range Video Reconstruction, CVPR 2023

Yixin Yang, Jin Han, Jinxiu Liang, Imari Sato, Boxin Shi.

Dependency

conda create -n HDRev python=3.7
conda activate HDRev
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
sudo apt install libopenexr-dev zlib1g-dev
conda install -c conda-forge openexr-python
pip install -r requirements.txt

Testing

We provide the testing code for both stacked voxel event data and text event data (t, x, y, p).

stacked voxel event data

one of the dataset example are shown in datasets/synthetic.

python test.py --dataroot datasets/synthetic

text event data

one of the dataset example are shown in datasets/real.

python test.py --dataroot datasets/real/ --test_on_txt

Training

Please ensure the pretraining path is correctly set in the model files before starting training.

event-to-image pretraining

python train.py --model E2IRecurrent --dataroot PATH_TO_DATA

LDR-to-image pretraining

python train.py --model L2LRecurrent --dataroot PATH_TO_DATA

event+LDR-to-HDR training

python train.py --model SEL2HDR_UnetRecurrent --dataroot PATH_TO_DATA

Contact

if you have any question, please send an email to yangyixin93@pku.edu.cn