Video Frame Interpolation via Generalized Deformable Convolution

Video frame interpolation aims at synthesizing intermediate frames from nearby source while maintaining spatial and temporal consistencies. The existing deep-learning-based video methods can be roughly divided into two categories: flow-based kernel-based methods. performance of is often jeopardized by the inaccuracy flow map estimation due to oversimplified motion models, that tends constrained rigidity kernel shape. To address these performance-limiting issues, a novel mechanism named generalized deformable convolution proposed, which effectively learn information in data-driven manner freely select sampling points space-time. We further develop new method based on this mechanism. Our extensive experiments demonstrate performs favorably against state-of-the-art, especially when dealing with complex motions. Code available https://github.com/zhshi0816/GDConvNet .