ERQA - a full-reference quality metric designed to analyze how good image and video restoration methods (SR, deblurring, denoising, etc) are restoring real details.

See our paper for details.

It is part of MSU Video Super Resolution Benchmark project.

Run pip install erqa and run it from command line or directly from Python code.

python -m erqa /path/to/target.png /path/to/gt.png

import erqa
import cv2

# Target and gt should be uint8 arrays of equal shape (H, W, 3) in BGR format
target = cv2.imread('/path/to/target.png')
gt = cv2.imread('/path/to/gt.png')

metric = erqa.ERQA()
v = metric(target, gt)

The ERQA metric analyzes how details were reconstructed in an image compared to ground-truth.

• ERQA = 1.0 means perfect restoration
• ERQA = 0.0 means worst restoration

Visualization of the metric shows underlying mask showing where image is distorted.

• Blue means there is a missing detail (False Negative)
• Red means there is a misplaced detail (False Positive)
• White means perfect details restoration (True Positive)
• Black means perfect background restoration (True Negative)

You can get source code up and running using following commands:

git clone https://github.com/msu-video-group/erqa
cd erqa
pip install -r requirements.txt

@conference{kirillova2022erqa,
	title        = {ERQA: Edge-restoration Quality Assessment for Video Super-Resolution},
	author       = {Anastasia Kirillova. and Eugene Lyapustin. and Anastasia Antsiferova. and Dmitry Vatolin.},
	year         = 2022,
	booktitle    = {Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 4: VISAPP,},
	publisher    = {SciTePress},
	pages        = {315--322},
	doi          = {10.5220/0010780900003124},
	isbn         = {978-989-758-555-5},
	organization = {INSTICC}
}