dadafeixia / deepqa-with-pytorch

Re-implement the work from "Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework"

DeepQA-with-Pytorch

This project aims to reimplement the work "Deep Learning of Human Visual Sensitivity in Image Quality Assessment Framework" for FR-IQA on PyTorch platform. The code has been trained and tested on LIVE and TID2013 database. The current performance is close to the claimed performance in the original paper.

File structure

This project folder should be included in a upper directory with the database data as following:

Project folder
└───DeepQA-with-Pytorch
│   │     README.md
│   │     train_LIVE.py
│   │     train_TID2013.py
│   └───datasets
│   └───models
│   └───snapshots
│   └───utils
│
└───data
│    └───LIVE_dataset
│    └───TID2013_dataset

How to use

Performance

The overall performance

	TID13-LCC	TID13-SROCC	LIVE-LCC	LIVE-SROCC
deep QA	0.947	0.939	0.982	0.981
This code	0.929	0.924	0.971	0.967

The "results" folder contains the training process and two examples for both LIVE and TID2013 datasets.

The training process:

LIVE

TID2013

Two examples from LIVE

Two examples from TID2013