Implementation of a SSIM loss function in MXNet Gluon. Implementation is based on Po Hsun Su's PyTorch implementation of SSIM.

SSIM stands for Structural Similarity Index and is a perceptual metric to measure similarity of two images. Commonly used loss functions such as L2 (Euclidean Distance) correlate poorly with image quality because they assume pixel-wise independance. For instance blurred images cause large perceptual but small L2 loss.

SSIM takes into account luminance, contrast and structure and is computed as follows:

To compare L2 loss and SSIM loss for a given image, execute python run.py myimage.jpg.