Hybrid Image

Hybrid images are images designed to convey multiple interpretations depending on viewing conditions, often exploiting differences in spatial frequency or other image components. Current research focuses on generating and utilizing hybrid images in diverse applications, including improving efficiency in 3D semantic segmentation (through combining 2D and 3D techniques), enhancing image processing for novel sensor technologies like HybridEVS cameras (using transformer-based architectures and coarse-to-fine approaches), and creating more realistic and controllable image generation through diffusion models and image decomposition. This work has implications for various fields, from autonomous driving (through improved scene understanding and labeling) to mobile imaging (via advanced image processing pipelines) and even fundamental neuroscience (by providing insights into human visual perception).