MIRNet

Learning Enriched Features for Real Image Restoration and Enhancement

Content

• MIRNet
• Getting Started
• Running
• Result
• References
• Citations

Getting Started

• Clone the repository

Prerequisites

• Tensorflow 2.2.0+
• Python 3.6+
• Keras 2.3.0
• PIL
• numpy

pip install -r requirements.txt

Running

Image Denoising

Training

• Get the dataset

  wget https://competitions.codalab.org/my/datasets/download/a26784fe-cf33-48c2-b61f-94b299dbc0f2
  

• Training

  python train_denoise.py
  

Testing

• Test

  python test_denoise.py
  

Usage

usage: train_denoise.py [-h] [--lr LR] [--gpu GPU]
                        [--grad_clip_norm GRAD_CLIP_NORM]
                        [--num_epochs NUM_EPOCHS]
                        [--train_batch_size TRAIN_BATCH_SIZE]
                        [--checkpoint_ep CHECKPOINT_EP]
                        [--checkpoint_filepath CHECKPOINT_FILEPATH]
                        [--num_rrg NUM_RRG] [--num_mrb NUM_MRB]
                        [--num_channels NUM_CHANNELS]

optional arguments:
                        -h, --help show this help message and exit
                        --lr LR
                        --gpu GPU
                        --grad_clip_norm GRAD_CLIP_NORM
                        --num_epochs NUM_EPOCHS
                        --train_batch_size TRAIN_BATCH_SIZE
                        --checkpoint_ep CHECKPOINT_EP
                        --checkpoint_filepath CHECKPOINT_FILEPATH
                        --num_rrg NUM_RRG
                        --num_mrb NUM_MRB
                        --num_channels NUM_CHANNELS

Download the weight here and put it to the weights/denoise folder.

usage: test_denoise.py [-h] [--test_path TEST_PATH] [--gpu GPU]
                       [--checkpoint_filepath CHECKPOINT_FILEPATH]
                       [--num_rrg NUM_RRG] [--num_mrb NUM_MRB]
                       [--num_channels NUM_CHANNELS]

optional arguments:
                        -h, --help  show this help message and exit
                        --test_path TEST_PATH
                        --gpu GPU
                        --checkpoint_filepath CHECKPOINT_FILEPATH
                        --num_rrg NUM_RRG
                        --num_mrb NUM_MRB
                        --num_channels NUM_CHANNELS

Image Super Resolution

Training

• Get the dataset here
• Training

  python train_super.py
  

Testing

• Test

  python test_super.py
  

Usage

usage: train_super.py [-h] [--lr LR] [--gpu GPU]
                      [--grad_clip_norm GRAD_CLIP_NORM]
                      [--num_epochs NUM_EPOCHS]
                      [--train_batch_size TRAIN_BATCH_SIZE]
                      [--checkpoint_ep CHECKPOINT_EP]
                      [--checkpoint_filepath CHECKPOINT_FILEPATH]
                      [--num_rrg NUM_RRG] [--num_mrb NUM_MRB] [--mode MODE]
                      [--scale_factor SCALE_FACTOR]

optional arguments:
                    -h, --help            show this help message and exit
                    --lr LR
                    --gpu GPU
                    --grad_clip_norm GRAD_CLIP_NORM
                    --num_epochs NUM_EPOCHS
                    --train_batch_size TRAIN_BATCH_SIZE
                    --checkpoint_ep CHECKPOINT_EP
                    --checkpoint_filepath CHECKPOINT_FILEPATH
                    --num_rrg NUM_RRG
                    --num_mrb NUM_MRB
                    --mode MODE
                    --scale_factor SCALE_FACTOR

Download the weight here and put it to the weights/super folder.

usage: test_super.py [-h] [--test_path TEST_PATH] [--gpu GPU]
                     [--checkpoint_filepath CHECKPOINT_FILEPATH]
                     [--num_rrg NUM_RRG] [--num_mrb NUM_MRB]
                     [--num_channels NUM_CHANNELS]
                     [--scale_factor SCALE_FACTOR]

optional arguments:
                    -h, --help            show this help message and exit
                    --test_path TEST_PATH
                    --gpu GPU
                    --checkpoint_filepath CHECKPOINT_FILEPATH
                    --num_rrg NUM_RRG
                    --num_mrb NUM_MRB
                    --num_channels NUM_CHANNELS
                    --scale_factor SCALE_FACTOR

Image Enhancement

Training

• Get the dataset here
• Training

  python train_delight.py
  

Testing

• Test

  python test_delight.py
  

Usage

usage: train_delight.py [-h] [--lr LR] [--gpu GPU]
                        [--grad_clip_norm GRAD_CLIP_NORM]
                        [--num_epochs NUM_EPOCHS]
                        [--train_batch_size TRAIN_BATCH_SIZE]
                        [--checkpoint_ep CHECKPOINT_EP]
                        [--checkpoint_filepath CHECKPOINT_FILEPATH]
                        [--num_rrg NUM_RRG] [--num_mrb NUM_MRB] [--mode MODE]

optional arguments:
                        -h, --help            show this help message and exit
                        --lr LR
                        --gpu GPU
                        --grad_clip_norm GRAD_CLIP_NORM
                        --num_epochs NUM_EPOCHS
                        --train_batch_size TRAIN_BATCH_SIZE
                        --checkpoint_ep CHECKPOINT_EP
                        --checkpoint_filepath CHECKPOINT_FILEPATH
                        --num_rrg NUM_RRG
                        --num_mrb NUM_MRB
                        --mode MODE

Download the weight here and put it to the weights/delight folder.

usage: test_delight.py [-h] [--test_path TEST_PATH] [--gpu GPU]
                       [--checkpoint_filepath CHECKPOINT_FILEPATH]
                       [--num_rrg NUM_RRG] [--num_mrb NUM_MRB]
                       [--num_channels NUM_CHANNELS]

optional arguments:
                        -h, --help            show this help message and exit
                        --test_path TEST_PATH
                        --gpu GPU
                        --checkpoint_filepath CHECKPOINT_FILEPATH
                        --num_rrg NUM_RRG
                        --num_mrb NUM_MRB

Result

Image Denoising

Input - Noisy	Output Denoised

Image Enhancement

Input - Lowlight	Output Delighted

Image Super Resolution

Input - LowRes	Output Bicubic	Output HighRes

License

This project is licensed under the MIT License - see the LICENSE file for details

References

• Training dataset
  • Denoise
  • Image Super Resolution
  • Image Enhancement

Citation

@misc{zamir2020learning,
      title={Learning Enriched Features for Real Image Restoration and Enhancement}, 
      author={Syed Waqas Zamir and Aditya Arora and Salman Khan and Munawar Hayat and Fahad Shahbaz Khan and Ming-Hsuan Yang and Ling Shao},
      year={2020},
      eprint={2003.06792},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgments

• Any ideas on updating or misunderstanding, please send me an email: vovantu.hust@gmail.com
• If you find this repo helpful, kindly give me a star.