SDAEDVSR: Efficient Video Super-Resolution With Stack Dense AutoEncoder

Recently video super-resolution has become important topics with the popularity of videos and a lot of end-toend neural network methods have been widely applied. Most related works focus on mapping relation whereas they ignore the relation between the high feature and the low feature. The low-resolution input video transforms into the high-resolution output video. Although many neural network methods achieve good performance, we use dense connectivity to consider that relation in order to make the video more clear. Therefore we propose an end-to-end autoencoder with a fusion of a dense network video super-resolution solution to strive to achieve realtime performance on the common mobile device in the condition of making maximum effort. We train our model to an efficient 4X video upscaling in the READS datasets. For the last time, we evaluated our solution and showed that it has significantly better performance and substantial speed improvement.