Wavelet-Based Network For High Dynamic Range Imaging

This is the official code for our paper "Wavelet-Based Network For High Dynamic Range Imaging" [CVIU 2023].

Requirements

• pytorch==1.4.0
• opencv-python
• scikit-image==0.17.2
• pywavelets==1.1.1

Datasets

Kalantari Dataset

Please download the kalantari dataset from this link.

RAW Dataset

Please download the kalantari dataset from this link (We are waiting for the approval to upload the dataset). The code for training the RAW dataset is on the raw branch.

Pretrained Model

Please download the pretained model from this link or you can use the following command:

wget https://github.com/TianhongDai/wavelet-hdr/releases/download/v1.0.0/model.pt.zip
unzip model.pt.zip

Instruction

1. train the network:

python train.py --cuda --use-bn --dataset-path <path-to-training-set> --testset-path <path-to-test-set>

2. continue training using the pre-saved checkpoint:

python train.py --cuda --use-bn --resume --last-ckpt-path <ckpt-path>

3. test the model and save HDR image:

python test.py --cuda --use-bn --save-path <model-path> --save-image

BibTex

To cite this code for publications - please use:

@article{dai2024wavelet,
  title={Wavelet-based Network for High Dynamic Range Imaging},
  author={Dai, Tianhong and Li, Wei and Cao, Xilei and Liu, Jianzhuang and Jia, Xu and Leonardis, Ales and Yan, Youliang and Yuan, Shanxin},
  journal={Computer Vision and Image Understanding},
  volume={238},
  pages={103881},
  year={2024},
  publisher={Elsevier}
}