RINO Framework Treats All Vision Tasks as RGB-to-RGB Image Editing

Researchers have proposed RGB In and RGB Out (RINO), a unified framework that represents all visual data — including depth maps, segmentation masks, and pose skeletons — as standard three-channel RGB images. Traditional computer vision systems require separate, task-specific architectures for different applications, making generalization difficult and costly. RINO reframes every vision task as a common RGB-to-RGB image editing problem, allowing a single model backbone to handle diverse tasks through a shared encoding and decoding structure. This design eliminates the need for task-specific fine-tuning, enabling zero-shot performance across multiple vision challenges. The approach draws a parallel to how language models operate over text, positioning RGB as a universal interface for visual understanding and generation.