DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks

This repository contains a reimplementation of the Automatic Photo Quality Enhancement using Deep Convolutional Networks project. The project aims to enhance the quality of photos using deep convolutional networks.

Environment Setup

To set up the environment, please follow these steps:

pip install -r ../requirements.txt

Training

To train the model, use the following command:

python train.py batch_size=<value> train_size=<value> lr=<value> epochs=<value> w_content=<value> w_color=<value> w_texture=<value> w_tv=<value> datadir=<path> vgg_pretrained=<path> eval_step=<value> model=<phone>

Training Arguments

The following arguments can be passed to the train.py script:

• batch_size=<value>: Set the batch size for training. Default: 50.
• train_size=<value>: Set the number of training samples. Default: 30000.
• lr=<value>: Set the learning rate. Default: 5e-4.
• epochs=<value>: Set the number of training epochs. Default: 20000.
• w_content=<value>: Set the weight for the content loss. Default: 10.
• w_color=<value>: Set the weight for the color loss. Default: 0.5.
• w_texture=<value>: Set the weight for the texture loss. Default: 1.
• w_tv=<value>: Set the weight for the total variation loss. Default: 2000.
• datadir=<path>: Specify the directory containing the DPED dataset. Default: dped/.
• vgg_pretrained=<path>: Specify the path to the VGG-19 pretrained model weights. Default: vgg_pretrained/imagenet-vgg-verydeep-19.mat.
• eval_step=<value>: Set the evaluation step. Default: 1000.
• model=<phone>: Specify the camera model to train. Available options: iphone, blackberry, sony.

Example usage:

python train.py batch_size=32 train_size=30000 lr=5e-4 epochs=20000 w_content=10 w_color=0.5 w_texture=1 w_tv=2000 datadir=dped/ vgg_pretrained=vgg_pretrained/imagenet-vgg-verydeep-19.mat eval_step=1000 model=iphone

Prediction

To predict and enhance the quality of photos, use the following command:

python predict.py model=<phone> datadir=<path> test_subset=<subset> iteration=<value> resolution=<value> use_gpu=<value>

Prediction Arguments

The following arguments can be passed to the predict.py script:

• model=<phone>: Specify the camera model to use for prediction. Available options: iphone, blackberry, sony, iphone_orig, blackberry_orig, sony_orig.
• datadir=<path>: Specify the directory containing the test images. Default: dped/.
• test_subset=<subset>: Specify the subset of test images to predict. Default: small.
• iteration=<value>: Specify the iteration number of the trained model checkpoint to use for prediction. Default: all.
• resolution=<value>: Specify the resolution of the output enhanced images. Available options: orig, `

Example usage:

python predict.py model=iphone datadir=test_img/ test_subset=full iteration=18000 resolution=orig use_gpu=true

Task Checklist:

1. Import Libraries - done
2. Initial Setting - done
3. Configure Data Loader - done
4. Define Generator - done
5. Define Discriminator - done
6. Training - done
7. Predict - done
8. Web (optional) - done
9. FINALIZATION (docstring, format, readme)

Citation

@inproceedings{ignatov2017dslr,
  title={DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks},
  author={Ignatov, Andrey and Kobyshev, Nikolay and Timofte, Radu and Vanhoey, Kenneth and Van Gool, Luc},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={3277--3285},
  year={2017}
}