A Generative Adversarial Network Based Tone Mapping Operator for 4K HDR Images

This is the repository for paper A Generative Adversarial Network Based Tone Mapping Operator for 4K HDR Images accepted to ICNC 2023.

Dataset

The dataset we constructed in Section 2.1 of our paper can be downloaded here.

Experiments

Training

Run the following code to train the network:

python -u esrgan.py --hdr_image_dir /path/to/HDR/ --sdr_image_dir /path/to/SDR/ --residual_blocks 23 --lr 1e-5 --warmup_epochs 300 --image_size 128 --batch_size 8 --psnr_decay_epoch 50 --decay_epoch 100 --n_epochs 600

Testing

Run the following code to evaluate the network. A pretrained model is provided here.

python -u infer.py --image_path /path/to/HDR/dir/ 
--output_dir /path/to/output/dir/ --checkpoint_model ./pretrained/generator.pth