weighted DCNN

Introduction

This is a tensorflow implementation of the network in "Bosse S, Maniry D, Müller K R, et al. Deep Neural Networks for No-Reference and Full-Reference Image Quality Assessment[J]. arXiv preprint arXiv:1612.01697, 2016." . We only reproduce the weighted FR network here. Refer to dmaniry/deepIQA for official source codes.

Dependencies

For the code in src

• Python 3.5
• TensorFlow 1.0
• Anaconda3
• Windows 10

For the code in src_tid2013_cluster

• Python 2.7
• TensorFlow 1.0
• CentOS 6.5

Results

We repeat 20 times on tid2013.

Note:

• the reference images and the epoch are indexed from 0.
• the metric includes MAE/SROCC/KROCC/PLCC/RMSE.

TODO

• complement the experiments on tid2013.