Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable

cn, https://srameo.github.io/projects/levd Input VRT † [26] Real-ESRGAN [46] MF2F [14] UDVD [40] Ours -RFCVD (sophisticated model) (sophisticated dataset) (sophisticated supervision) (sophisticated supervision) Figure 1. In real-world videos, noise can manifest in vastly different ways, causing supervised methods to fail when the input is too far outside the training distribution. Footage provided by Robert Kjettrup, face pixelated after inference. We denote retrained models with a †.