DeepDC: Deep Distance Correlation as a Perceptual Image Quality Evaluator

This is the repository of paper DeepDC: Deep Distance Correlation as a Perceptual Image Quality Evaluator.

Highlights:

• A novel FR-IQA model that fully utilizes the texture-sensitive of pre-trained deep neural network (DNN) features, which computes distance correlation in the deep feature domain
• The model is exclusively based on the features of the pre-trained DNNs and does not rely on fine-tuning with mean opinion scores (MOSs)
• Extensive experiments achieve superior performance on five standard IQA datasets, one perceptual similarity dataset, two texture similarity datasets, and one geometric transformation dataset
• It can be employed as an objective function in texture synthesis and neural style transfer

====== PyTorch Implementation ======

Installation:

• pip install DeepDC-PyTorch

Requirements:

• Python >= 3.6
• PyTorch >= 1.0

Usage:

from DeepDC_PyTorch import DeepDC
model = DeepDC()
# calculate DeepDC between X, Y (a batch of RGB images, data range: 0~1) 
deepdc_score = model(X, Y)

or

git clone https://github.com/h4nwei/DeepDC
cd DeepDC_PyTorch
python DeepDC.py --ref <ref_path> --dist <dist_path>

Reference

• R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 586–595.
• K. Ding, K. Ma, S. Wang, and E. P. Simoncelli, “Image quality assessment: Unifying structure and texture similarity,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 5, pp. 2567–2581, 2020.
• I. Kligvasser, T. Shaham, Y. Bahat, and T. Michaeli, “Deep self dissimilarities as powerful visual fingerprints,” in Neural Information Processing Systems, 2021, pp. 3939–3951.

Citation

@article{zhu2023DeepDC,
title={DeepDC: Deep Distance Correlation as a Perceptual Image Quality Evaluator},
author={Zhu, Hanwei and Chen, Baoliang and Zhu, Lingyu and Wang, Shiqi and Lin, Weisi},
journal={CoRR},
volume = {abs/2211.04927v2},
year={2023},
url = {https://arxiv.org/pdf/2211.04927v2.pdf}
}