Semantic Segmentation of Outdoor Scenes using Adversarial training of a Efficient Net Based Generator and Patch-GAN based discriminator

This project shows an Image Semantic Segmentation framework, trained using Adversarial training.

The Generator was based on the Efficient-Net architecture, whereas the discriminator featured a Patch-GAN Architecture.

The resulting Neural Network was fast (>20 fps on a Mobile RTX 2060), and accurate.

Sample Efficient Net b4

Sample PatchGAN Architecture