Efficient lightweight network for video super-resolution

Recently, video super-resolution has achieved an outstanding performance. However, many existing methods to solve video super-resolution usually make use of complex strategies, such as explicit optical flow, deformable convolution, which increase complexity and computation. In this paper, we propose a lightweight network for video super-resolution, namely Efficient Lightweight Network for Video Super-Resolution (ELNVSR). We design a Multi-group Block extracting long-distance spatial information to construct a lightweight Bidirection Alignment Module which is implicitly capable of fusing and propagating spatial-temporal information in a bidirectional way. Meanwhile, a Multi-scale Pyramid Block is built as a lightweight reconstruction module to extract different levels of information layer by layer. Comprehensive experiments are conducted on public benchmarks. The results demonstrate a promising performance with fewer parameters.