Video frame interpolation aims to generate intermediate frames in a video to showcase finer details. However, most methods are only trained and tested on low-resolution datasets, lacking research on 4K video frame interpolation problems. This limitation makes it challenging to handle high-frame-rate video processing in real-world scenarios. In this paper, we propose a 4K video dataset at 120 fps, named UHD4K120FPS, which contains large motion. We also propose a novel framework for solving the 4K video frame interpolation task, based on a multi-scale pyramid network structure. We introduce self-attention to capture long-range dependencies and self-similarities in pixel space, which overcomes the limitations of convolutional operations. To reduce computational cost, we use a simple mapping-based approach to lighten self-attention, while still allowing for content-aware aggregation weights. Through extensive quantitative and qualitative experiments, we demonstrate the excellent performance achieved by our proposed model on the UHD4K120FPS dataset, as well as illustrate the effectiveness of our method for 4K video frame interpolation. In addition, we evaluate the robustness of the model on low-resolution benchmark datasets.

Traditional video frame interpolation methods based on deep convolutional neural networks face challenges in handling large motions. Their performance is limited by the fact that convolutional operations cannot directly integrate the rich temporal and spatial information of inter-frame pixels, and these methods rely heavily on additional inputs such as optical flow to model motion. To address this issue, we develop a novel framework for video frame interpolation that uses Transformer to efficiently model the long-range similarity of inter-frame pixels. Furthermore, to effectively aggregate spatio-temporal features, we design a novel attention mechanism divided into temporal attention and spatial attention. Specifically, spatial attention is used to aggregate intra-frame information, integrating both attention and convolution paradigms through the simple mapping approach. Temporal attention is used to model the similarity of pixels on the timeline. This design achieves parallel processing of these two types of information without extra computational cost, aggregating information in the space–time dimension. In addition, we introduce a context extraction network and multi-scale prediction frame synthesis network to further optimize the performance of the Transformer. Our method and state-of-the-art methods are extensively quantitatively and qualitatively experimented on various benchmark datasets. On the Vimeo90K and UCF101 datasets, our model achieves improvements of 0.09 dB and 0.01 dB in the PSNR metrics over UPR-Net-large, respectively. On the Vimeo90K dataset, our model outperforms FLAVR by 0.07 dB, with only 40.56% of its parameters. The qualitative results show that for complex and large-motion scenes, our method generates sharper and more realistic edges and details.