Table 9 Summary of Quantitative Evaluation Metrics for Image Quality
From: Improving generalization of polyp detection via conditional StyleGAN augmented training
Metric | Core Principle | Interpretation |
|---|---|---|
FID | Measures the Wasserstein 2 distance between the distributions of deep features (from InceptionV3) of real and generated images. | Lower is better |
IS | Measures the KL divergence between conditional and marginal class distributions of generated images to assess clarity and diversity. | Higher is better |
LPIPS | Computes the distance between two images in a learned deep feature space, correlating with human perceptual judgment. | Lower is better |
