grseb9s / srcnnkit

CoreML and Keras implementation of Super-Resolution Convolutional Neural Network (SRCNN)

SRCNNKit

Implementation of Super Resolution (SR) with CoreML and Swift. You can use SR method in your app using SRCNNKit UIImageView extension.

For details, see the following presentaion:

https://speakerdeck.com/kenmaz/super-resolution-with-coreml-at-try-swift-tokyo-2018

Pre trained model

Sorry, this project dosen't contain *.mlmodel yet. You should train your own model and import SRCNN.mlmodel to your project.

or use this fork: https://github.com/kenmaz/SRCNNKit

iOS

Usage

import SRCNNKit

let imageView: UIImageView = ...
let image: UIImage = ...

imageView.setSRImage(image)

Install

• Copy sources to your project.
• CocoaPods and Carthage will be supported soon.

Requirements

• iOS11
• Xcode9.x

Run sample project

• Copy your SRCNN.mlmodel to model directory
• Run following command:

git submodule init
git submodule update

• Open SRCNN-ios/SRCNN-ios.xcodeproj and Run

Train Your own model

Requirements

• Python 3.0+
• see script/packages.txt

Convert Training Data

cd script
python3 convert.py <original train image dir> <train data dir>
python3 convert.py <original validation image dir> <validation data dir>

Training

python3 train.py <tf log dir> <model output dir> <train data dir> <validation data dir>

Plot Model Image

python plot.py <.h5 model path> <output dir>

Convert Keras to CoreML Model

python3 coreml_convert.py <h5 mode path> <output dir>

Validate CoreML Model

python3 coreml_predict.py <mlmodel path> <input patch image path> <output patch image path>

Dependencies

https://github.com/hollance/CoreMLHelpers

Licence

SRCNNKit is released under the MIT license. See LICENSE for details.

Copyright © 2018 DeNA Co., Ltd. All rights reserved.