Learning Spatial and Spatio-Temporal Pixel Aggregations for Image and Video Denoising