ST-MFNet: A Spatio-Temporal Multi-Flow Network for Frame Interpolation

TL;DR

This approach has been comprehensively evaluated - compared with fourteen state-of-the-art VFI algorithms - clearly demonstrating that ST-MFNet consistently outperforms these benchmarks on var-ied and representative test datasets, with significant gains up to 1.09dB in PSNR for cases including large motions and dynamic textures.

Abstract

Video frame interpolation (VFI) is currently a very active research topic, with applications spanning computer vision, post production and video encoding. VFI can be extremely challenging, particularly in sequences containing large motions, occlusions or dynamic textures, where existing approaches fail to offer perceptually robust interpolation performance. In this context, we present a novel deep learning based VFI method, ST-MFNet, based on a Spatio-Temporal Multi-Flow architecture. ST-MFNet employs a new multi-scale multi-flow predictor to estimate many-to-one intermediate flows, which are combined with conventional one-to-one optical flows to capture both large and complex motions. In order to enhance interpolation performance for various textures, a 3D CNN is also employed to model the content dynamics over an extended temporal window. Moreover, ST-MFNet has been trained within an ST-GAN framework, which was originally developed for texture synthesis, with the aim of further improving perceptual interpolation quality. Our approach has been comprehensively evaluated -- compared with fourteen state-of-the-art VFI algorithms -- clearly demonstrating that ST-MFNet consistently outperforms these benchmarks on varied and representative test datasets, with significant gains up to 1.09dB in PSNR for cases including large motions and dynamic textures. Project page: https://danielism97.github.io/ST-MFNet.

Figures (10)

• Figure 1: High-level architecture of ST-MFNet, which employs a two-stage workflow to interpolate an intermediate frame.
• Figure 2: Illustration of the MIFNet. (a) The overall architecture of MIFNet, with a U-Net style backbone and multi-flow estimation heads at three scales. (b) The convolutional layers inside the multi-flow head at each scale.
• Figure 3: Illustration of the MSResNext block, which consists of two ResNext branches with different kernel sizes, followed by a channel attention module.
• Figure 4: The architecture of the Texture Enhancement Network.
• Figure 5: Qualitative results interpolated by different variants of our method. Here "Overlay" means the overlaid adjacent frames. Figures (a)-(d): w/ MIFNet vs w/o MIFNet; figures (e)-(h): w/ BLFNet vs w/o BLFNet; figures (i)-(j): UMSResNext vs U-Net; figures (m)-(p): w/ TENet vs w/o TENet; figures (q)-(v): comparison of different GANs.