MOTION BLURING AND DEPTH OF DEPTH RECONSTRUCTION THROUGH TIME-STABLE NEURONAL NETWORKS

摘要

A structure of a neural network, namely a distorted external recurrent neural network, is disclosed for the reconstruction of images with synthesized effects. The effects can include motion blur, depth of field reconstruction (e.g. simulation of lens effects) and / or anti-aliasing (e.g. removal of artifacts caused by a sampling frequency). The distorted external recurrent neural network is not recurrent on every layer within the neural network. Instead, the external state output from the last layer of the neural network is distorted and provided as part of the input to the neural network for the next image in a sequence of images. In contrast, in a conventional recurrent neural network, a hidden state generated on each layer is provided as a feedback input for the generating layer. The neural network can be implemented at least partially on a processor. In one embodiment, the neural network is implemented on at least one parallel processing unit.