Scaling up GANs for Text-to-Image Synthesis
Introduces GigaGAN, a new GAN architecture that far exceeds the capacity of the StyleGAN architecture, offering faster inference time and the ability to synthesize high-resolution images, making GANs a viable option for text-to-image synthesis.
Implementing GigaGAN can help improve the speed and quality of text-to-image synthesis for businesses in industries such as e-commerce and advertising.
Cherry-Picking with Reinforcement Learning
Presents CherryBot, an RL system that surpasses human reactiveness for some dynamic grasping tasks, demonstrating continual improvement through real-world interaction, and generalizability to varying object shapes and dynamics.
Implementing CherryBot can help businesses in industries such as manufacturing, construction, and disaster recovery by improving the efficiency and precision of robotic manipulation tasks.