PSNR Calculation MATLAB Code for PIRM-SR Challenge
The PIRM-SR challenge aims to compare and rank perceptual single-image super-resolution methods. In terms of perceptual quality, state-of-the-art methods often perform poorly when evaluated with 'simple' distortion metrics like PSNR and SSIM. Hence, in contrast to previous challenges, the evaluation and ranking will focus on perceptual quality, adopting a unified approach that combines algorithm accuracy with perceptual quality. This allows perceptual-driven methods to compete with those designed to maximize PSNR.
To self-verify your method, use this MATLAB code to compute RMSE and perceptual scores for your output on a self-validation set. Here's how to quickly get started:
- Copy your output to the
your_results
folder in the base directory. - Copy only the HR images to the
self_validation_HR
folder. - Download and extract the SR-Metric toolbox into the
utils/sr-metric-master
folder. - Run the
evaluate_results.m
script.
Troubleshooting
Depending on your operating system, you might need to recompile the MEX files in the matlabPyrTools toolbox. If that's the case, follow these steps:
- Navigate to utils/sr-metric-master/external/matlabPyrTo and recompile the MEX files.