/deepBoosting

Deep Boosting for Image Denoising in ECCV 2018 and its Real-world Extension in IEEE Transactions on Pattern Analysis and Machine Intelligence

Primary LanguagePythonMIT LicenseMIT

Deep Boosting for Image Denoising

Official implementation for Deep Boosting Framework introduced in the following papers:
Chang Chen, Zhiwei Xiong, Xinmei Tian, Feng Wu. Deep Boosting for Image Denoising. In ECCV 2018.
Chang Chen, Zhiwei Xiong, Xinmei Tian, Zheng-Jun Zha, Feng Wu. Real-world Image Denoising with Deep Boosting. IEEE Transactions on Pattern Analysis and Machine Intelligence, in press.

Requirements

Anaconda>=4.2.0 (Python 3.5)
TensorFlow==1.4.0
Matlab Engine (Python Interface)

Train the model

Usage example to train/evaluate a new model

cd train && cat train400.tfrecord.tar.gz.* | tar -xzv
python train.py && python inference.py

Test the pre-trained models

Usage example to non-blind gray-level Gaussian image denoising

cd dn-nonblind-gray && python eval.py

Usage example to blind gray-level Gaussian image denoising

cd dn-blind-gray && python eval.py

Usage example to blind color Gaussian image denoising

cd dn-blind-color && python eval.py

Usage example to JPEG image deblocking

cd db-gray && python eval.py

Usage example to in-domain real-world image denoising

cd dn-real-in && python eval.py

Usage example to cross-domain real-world image denoising

cd dn-real-cross && python eval.py

RID Dataset

To access Real-world Image Denoising (RID) dataset for training and validation
Download RID.tar.gz.0~5 from
http://pan.bitahub.com/index.php?mod=shares&
sid=eTJ2bFFQR3BzTm5FTGdjdXFBUnl2Y3htd3puWjFwRDc4SU9vSXc

cat RID.tar.gz.* | tar -xzv

To access Set60 benchmark for testing

cd datas/Set60