Omnidirectional Image Super Resolution

Omnidirectional image super-resolution (ODISR) aims to enhance the resolution of 360° images, crucial for applications like virtual and augmented reality. Current research focuses on addressing the geometric distortions inherent in projecting spherical images onto planar formats, employing techniques like transformers with distortion-aware mechanisms and novel sampling strategies for viewport rendering. These advancements improve image quality and realism, particularly in head-mounted display applications, by leveraging both deep learning models and diffusion-based approaches, even achieving zero-shot super-resolution. The resulting improvements in visual fidelity are significant for immersive experiences and related visual tasks.