Photorealism evaluation metrics
One of the main issues within this subject matter is quantitatively assessing the photorealism of the generated synthetic images. In this section, we will explore the main metrics usually used. We will explore the following:
- Structural Similarity Index Measure (SSIM)
- Learned Perceptual Image Patch Similarity (LPIPS)
- Expert evaluation
Structural Similarity Index Measure (SSIM)
SSIM is one of the most widely used metrics to measure the structural similarity between two images. It was first introduced in the paper titled Image quality assessment: from error visibility to structural similarity (https://ieeexplore.ieee.org/document/1284395). The SSIM metric does not compare individual pixels of the two images. However, it considers a group of pixels assuming that spatially close pixels have inter-dependencies. These dependencies can be linked to the actual structure of objects that were captured and presented by the given images...
