Tech Session Advanced Tech, AI

Wallpaper Translation using Generative AI

Recent progress in text-to-image and image-to-image generative AI using diffusion models allows us to deploy the AI model on devices. Our model translates wallpaper on real-time weather and time information, providing users with informative and emotionally engaging experiences. This task involves resolving two conflicting objectives: accurately reflecting weather and time conditions (in line with developers' preferences) while maintaining the structural integrity of target objects. To address this challenge, we have developed a novel Seg-scale approach that leverages external segmentation results as a mask for classifier-free guidance, enabling us to seamlessly incorporate weather and time conditions into the background while preserving object structures without incurring additional computational costs. Second, we fine-tune the conditions to meet our preferences.

Speakers

Juyong Song

Samsung Electronics

Preferences Submitted

You have successfully updated your cookie preferences.