MEMC-Net: Motion Estimation and Motion Compensation Driven Neural Network for Video Interpolation and Enhancement

Motion estimation (ME) and motion compensation (MC) have been widely used for classical video frame interpolation systems over the past decades. Recently, a number of data-driven frame interpolation methods based on convolutional neural networks have been proposed. However, existing learning based methods typically estimate either flow or compensation kernels, thereby limiting performance on both computational efficiency and interpolation accuracy. In this work, we propose a motion estimation and compensation driven neural network for video frame interpolation. A novel adaptive warping layer is developed to integrate both optical flow and interpolation kernels to synthesize target frame pixels. This layer is fully differentiable such that both the flow and kernel estimation networks can be optimized jointly. The proposed model benefits from the advantages of motion estimation and compensation methods without using hand-crafted features. Compared to existing methods, our approach is computationally efficient and able to generate more visually appealing results. Furthermore, the proposed MEMC-Net architecture can be seamlessly adapted to several video enhancement tasks, e.g., super-resolution, denoising, and deblocking. Extensive quantitative and qualitative evaluations demonstrate that the proposed method performs favorably against the state-of-the-art video frame interpolation and enhancement algorithms on a wide range of datasets.

1 Introduction

Video frame interpolation aims to synthesize non-existent frames between original input frames, which has been applied to numerous applications such as video frame rate conversion [1], novel view synthesis [2], and frame recovery in video streaming [3], to name a few. Conventional approaches [4], [5] are generally based on motion estimation and motion compensation (MEMC), and have been widely used in various display devices [6]. A few deep learning based frame interpolation approaches [7], [8] have been developed to address this classical topic. In this paper, we analyze the MEMC-based and learning-based approaches of video frame interpolation and exploit the merits of both paradigms to propose a high-quality frame interpolation processing algorithm.