SSIM, Structural Similarity, structural similarity. It is also an index to measure the similarity of two pictures. The basic principle of that natural images are considered highly structured, that is, adjacent pixels are strongly correlated with each other. The human visual system has a strong ability to understand and extract information from images, so when measuring image quality, structural distortion is crucial to take into consideration. When calculating the two pictures For structural similarity, local sliding window processing is used, usually, small blocks of NxN, to calculate the structural similarity in the sliding window, sliding the sliding window in pixels each time until the local structure of each position in the entire picture is similar After all the calculations are completed, averaging all the local structural similarities is the structural similarity of the two pictures. The equation is illustrated in the SSIM loss, Section 3.4.6.4.