Temporal restoration applications. (Top) Event to video generation models convert a spatio-temporal stream of events with microsecond temporal resolution into a high-quality video, which is typically monochromatic (however, colored outputs also exist through generative coloring). This enables applications such as high-framerate videos and high dynamic range capture. Usually, these models take in a 3D tensor of events along with the feedback from the previous predicted frame to generate the next video frame. (Bottom) Frame interpolation models leverage the high temporal resolution of event streams to estimate non-linear motion information between the frames and insert latent frames between two consecutive frames. As we discuss below, these models can be trained with full supervision, weak supervision or unsupervised methods. The general approach for such models follows processing the RGB and event streams separately and then fusing the information, followed by refinement