Fig. 13: Validation of conditional generation.
From: Improving generalization of polyp detection via conditional StyleGAN augmented training
This grid provides a direct visual comparison between real endoscopic images (left column) and synthetic images (middle, right columns) generated by providing specific class labels to the generator. The examples cover the three morphological categories used for conditioning: (top row) tubulovillous adenoma, (middle row) pedunculated polyp, and (bottom row) sessile polyp. This qualitatively demonstrates the model’s fidelity and its ability to control the generated morphology based on the input condition.
