Deep Recursive HDRI in Pytorch

paper

We provide PyTorch implementations for GAN-based mutliple exposure stack generation.

• Deep recursive HDRI

General

If you use the code for your research work, please cite our papers.

@inproceedings{lee2018deep,
  title={Deep recursive hdri: Inverse tone mapping using generative adversarial networks},
  author={Lee, Siyeong and Hwan An, Gwon and Kang, Suk-Ju},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={596--611},
  year={2018}
}

Model inference

• Conda environment

conda create -n hdr python=3.6
conda activate hdr
conda install -c anaconda mkl
conda install pytorch==1.0.0 torchvision==0.2.1 cuda100 -c pytorch

• install requirements.txt

pip install -r requirements.txt

• Please download two model weights below and organize the downloaded files as follows:

DeepRecursive_HDRI
├──Result
    └──model
       ├── HDRGAN_stopdown_G_param_ch3_batch1_epoch20_lr0.0002.pkl
       └── HDRGAN_stopup_G_param_ch3_batch1_epoch20_lr0.0002.pkl

• Prepare your test images

DeepRecursive_HDRI
├──input
   ├── t10.png 
   ├── t11.png

• Run the pretrained model

python test.py --test_dataset './input'

• output

DeepRecursive_HDRI
├──Result
   ├── t10 (multi exposure stack)
   ├── t11 (multi exposure stack)

Note: We used the HDR Toolbox implementation of [Debevec and Malik 1997] to generate the results in our paper.

• see https://github.com/banterle/HDR_Toolbox

Model weight

Model Name	model weight
Deep Recursive HDRI	stopdown stopup

License

Copyright (c) 2020, Siyeong Lee. All rights reserved.

The code is distributed under a BSD license. See LICENSE for information.