Deep Recursive HDRI in Pytorch

paper

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

• [x] 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

License

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

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