DRLIE

This is official Pytorch implementation of "DRLIE: Flexible Low-Light Image Enhancement via Disentangled Representations"

Framework

A typical schematic of our proposed DRLIE.

Schematic of the proposed custom illumination guided low-light image enhancement algorithm based on disentangle representations.

Framework of disentanglement model for multiexposure images.

Network architecture of the content encoder, attribute encoder, and generator.

Coding

Recommended Environment

tensorflow-gpu 1.14.0
scipy 1.2.0
numpy 1.19.2
opencv 3.4.2

To Train

Please load the training dataset from here and place it in this project.

Then, training the disentanglement model by entering the following code

CUDA_VISIBLE_DEVICES=0,1 python main.py --phase train --dataset over2under

Moreover, you also could place the underexposed images from the MEF dataset in ./dataset/over2under/trainA and put the overexposed images in ./dataset/over2under/trainB. Then, rewrite the dataloader to load your dataset and retrain your model.

Note: the training of the disentanglement model is extremely unstable and may not yields an excellent model.

To Test

CUDA_VISIBLE_DEVICES=0,1 python main.py --phase guide --dataset AGLIE --guide_num 129 --batch_size 1 --direction a2b

The test images from the AGLIE datset please put in ./dataset/AGLIE/ and the guided images please put in ./guide/ . You can modify the relevant configureation according to our producedures.

Experiment Results

Visual results of different low-light image enhancement methods on the AGLIE dataset.

Visual results of different low-light image enhancement methods on the MEF dataset.

Visual results of different low-light image enhancement methods on the LOL dataset.

Visual results of different low-light image enhancement methods on the VV dataset.

Example of custom illumination adjustment with specific exposure levels.

Visual results of controllable illumination manipulation with multiexposure guide images.

Some typical examples of flexible illumination adjustment, guided by the MIT-Adobe FiveK dataset.

Citation

@article{Tang2022DRLIE,
  author={Tang, Linfeng and Ma, Jiayi and Zhang, Hao and Guo, Xiaojie},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={DRLIE: Flexible Low-Light Image Enhancement via Disentangled Representations}, 
  year={2022},
  volume={},
  number={},
  pages={1-14},
  doi={10.1109/TNNLS.2022.3190880}}

Acknowledgement

The codes are heavily based on DRIT-Tensorflow. Please also follow their licenses. Thanks for their awesome works.

If you have any questions, please contact linfeng0419@gmail.com.

qingchen2333/DRLIE