Fig. 7: A multi-scale generative network for fluorescein angiography (FA) synthesis from color fundus.

The model comprises of two generators and two discriminators for taking in images of different resolutions and scales. There is a feature fusion between the coarse and fine generators, which extracts both global and local features. The discriminators take in pairs of fundus and FA images and dictates if the image pairs are real or fake. The generator only other hand synthesizes FA from retinal fundus images at two different scales.