Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis

A black and white photo of a dog with big eyes. A steaming plate of classic fried rice with vegetables, eggs. A photo of a colorful sports car, with mountain background. Oil painting of a flower garden, distant cottage, soft sunlight. Sunrise over the clouds in the mountains. A pencil sketch of a woman with deep, almond-shaped eyes in. A coral reef with starfish, sea anemones and fish. Figure 1. Example results at 512×512 resolution synthesized by our proposed Aurora, a large-scale GAN-based text-to-image generator.