ECCV'18: Deep High Dynamic Range Imaging with Large Foreground Motions

This is the implementation for Deep High Dynamic Range Imaging with Large Foreground Motions, Shangzhe Wu, Jiarui Xu, Yu-Wing Tai, Chi-Keung Tang, in ECCV, 2018. More results can be found on our project page.

Get Started

Prerequisites

• Python 3.5
• Tensorflow 1.4.0
• OpenCV 3.4 (from conda-forge)
• Photomatix for tonemapping

Setup

• Clone this repo:

git clone https://github.com/elliottwu/DeepHDR.git
cd DeepHDR

• Download pretrained model: (~60MB)

sh download_pretrained.sh

Demo

sh test.sh

Tonemapping (post-processing)

Generated HDR images are in .hdr format, which may not be properly displayed in your image viewer directly. You may use Photomatix for tonemapping:

• Download Photomatix free trial, which won't expire.
• Load the generated .hdr file in Photomatix.
• Adjust the parameter settings. You may refer to pre-defined styles, such as Detailed and Painterly2.
• Save your final image in .tif or .jpg.

Train

• Download Kalantari's dataset: (~8GB)

cd dataset
sh download_dataset.sh
cd ..

• Prepare TFRecord: (this takes ~10 minutes)

cd dataset
python convert_to_tfrecord.py
cd ..

• Start training:

sh train.sh

• To monitor training using Tensorboard, copy the following to your terminal and open localhost:8888 in your browser

tensorboard --logdir=logs --port=8888

Citation

@InProceedings{Wu_2018_ECCV,
  author = {Wu, Shangzhe and Xu, Jiarui and Tai, Yu-Wing and Tang, Chi-Keung},
  title = {Deep High Dynamic Range Imaging with Large Foreground Motions},
  booktitle = {The European Conference on Computer Vision (ECCV)},
  month = {September},
  year = {2018}
}