Video Frame Interpolation Based on Lightweight Convolutional Unit and Three-scale Encoder
Weijing Liu, Xiaohui Yang, Zhiquan Feng, Tao Xu, Qingbei Guo · 2023
Video frame interpolation (VFI) achieves temporal super-resolution by synthesizing intermediate frame between two original adjacent frames. This paper proposes a lightweight VFI network based on lightweight convolutional unit and three-scale encoder. We first introduce a three-scale encoding-decoding structure with two-level attention cascades to represent the multi-scale motion information. Then, recurrent convolutional layer (RCL) and residual operation are adopted to design the recurrent residual convolutional unit (RRCU) to refine the three-scale structure. Finally, we propose a lightweight unit by combining the depth separable convolution and RRCU, and introduce the local lightweight idea to reduce the model parameters. Experimental results show that the proposed method achieves a better performance against the state-of-the-art methods with fewer parameters.