This repository contains code and pretrained models for our paper: Self-Gated Memory Recurrent Network for Efficient Scalable HDR Deghosting accepted to IEEE Transactions on Computational Imaging.
It has been tested on RTX 6000 with Tensorflow 2.3.4.
Getting base image:
$ docker pull tensorflow/tensorflow:2.3.4-gpuRunning base image:
$ docker run --rm -it tensorflow/tensorflow:2.3.4-gpu bashInstalling dependencies:
(docker)# apt update
(docker)# apt install -y ffmpeg libsm6 libxext6 libxrender-dev
(docker)# pip install opencv-pythonThe Kalantari dataset (SIG17) can be downloaded here, and the Prabhakar dataset (ICCP19) can be downloaded here.
Download the required dataset and extract it in the dataset folder.
To view all training options, run
$ python main.py --helpTo train a Bidirectional SGM model with the default configuration, run
$ python main.py --rtxTo evaluate the pretrained model on the Kalantari17 dataset, run
$ python val.py --rtx --weights pretrained_weights/UCSD/bidirectional.sgm/bidirectional.sgm.tfWhen citing this work, you should use the following Bibtex:
@ARTICLE{9540317,
author={Kathirvel, Ram Prabhakar and Agrawal, Susmit and Radhakrishnan, Venkatesh Babu},
journal={IEEE Transactions on Computational Imaging},
title={Self-Gated Memory Recurrent Network for Efficient Scalable HDR Deghosting},
year={2021},
volume={},
number={},
pages={1-1},
doi={10.1109/TCI.2021.3112920}
}
