deep_battery_saver's Introduction

Deep Battery Saver: End-to-end Learning for Power Constrained Contrast Enhancement

The details about our deep battery saver can be found in our paper.

License + Attribution

This code is licensed under CC BY-NC-SA 4.0. Commercial usage is not permitted. If you use this code in a scientific publication, please cite the following paper:

@ARTICLE{yin_tmm20_pcce,
author={J. -L. {Yin} and B. -H. {Chen} and Y. -T. {Peng} and C. -C. {Tsai}},
journal={IEEE Transactions on Multimedia}, 
title={Deep Battery Saver: End-to-End Learning for Power Constrained Contrast Enhancement}, 
year={2021},
volume={23},
number={},
pages={1049-1059},
doi={10.1109/TMM.2020.2992962}}

Dependencies

Python 3
Anaconda 3
Tensorflow >= 2.0.0 (CUDA version >= 10.0 if installing with CUDA. More details)
Python packages: You can install the required libraries by the command conda DeepBatterySaver -n recreated_env --file requirements.txt. We checked this code on cuda-10.0 and cudnn-7.6.5.

It was tested and runs under the following OSs:

Windows 10
Ubuntu 16.04/18.04

Might work under others, but didn't get to test any other OSs just yet.

Usage

Clone this github repo.

git clone https://github.com/bigmms/deep_battery_saver

Place your testing images in ./test folder. (There are several sample images).
Then, cd to deep_battery_saver and run following command for evaluation:

python main.py --source_path ./test/pills.jpg

To run with different settings, add --num_epochs, --power_level, --learning_rate, --source_path, --target_path as you need.
The results are in ./results folder.

Results

deep_battery_saver's People

Contributors

Stargazers

Watchers

deep_battery_saver's Issues

Paper architecture and github architecture

Hi,
My Name Is Emmanuel Sampaio.

I am contacting you because I am currently comparing different approaches to perform power-constrained contrast enhancement (PCCE). Among the different models I tested, yours proved to be really interesting and innovative, and for this reason I would like to check its performance in comparison with other deep and non-deep approaches.

Given this context, I would like to ask you if the implementation presented in this GitHub corresponds to the implementation discussed in this paper:
https://ieeexplore.ieee.org/document/9089316/citations#citations

I'm asking this because I noticed some differences:

Residual blocks: the code uses a function to add block's input and output, while the paper mentioned a simple concatenation.
The output function: the paper not mentioned the use of a Tanh.
The parameter $\lambda$ from the power loss is constantly set equal to 30.

Thank you for your attention,
Emmanuel Sampaio

Recommend Projects