A Fast 4K Video Frame Interpolation based on StepWise Optical Flow Computation and Video Spatial Interpolation

Recently, with the generalization of ultra-high-resolution displays and the beginning of terrestrial UHD TV, the demand for 4K video is spreading. However, the supply for 4k contents is still insufficient compared to the demand. To create the 4K video with high temporal resolution, this paper proposes video frame interpolation method for 4K video. Video frame interpolation with high spatial resolution requires a large receptive field of the convolution kernel due to its resolution and also consumes a lot of GPU memory and computation. To solve this problem, in this paper, a frame interpolation method that combines a stepwise optical flow computation method and video spatial interpolation is proposed. Experiment results show that the proposed method is not only objectively and subjectively superior to the existing method, but also significantly reduces the GPU memory and computational amount compared to the existing one.