Spatial-Temporal Residual Aggregation for High Resolution Video Inpainting

TL;DR

STRA-Net is a novel framework that enhances high-resolution video inpainting by combining low-resolution inpainting with residual aggregation to improve temporal consistency and visual quality.

Contribution

The paper introduces STRA-Net, a new spatial-temporal residual aggregation approach that enables high-resolution video inpainting without heavy memory use, outperforming existing methods.

Findings

Produces more temporally-coherent inpainted videos

Achieves higher visual quality in high-resolution videos

Outperforms state-of-the-art methods in evaluations

Abstract

Recent learning-based inpainting algorithms have achieved compelling results for completing missing regions after removing undesired objects in videos. To maintain the temporal consistency among the frames, 3D spatial and temporal operations are often heavily used in the deep networks. However, these methods usually suffer from memory constraints and can only handle low resolution videos. We propose STRA-Net, a novel spatial-temporal residual aggregation framework for high resolution video inpainting. The key idea is to first learn and apply a spatial and temporal inpainting network on the downsampled low resolution videos. Then, we refine the low resolution results by aggregating the learned spatial and temporal image residuals (details) to the upsampled inpainted frames. Both the quantitative and qualitative evaluations show that we can produce more temporal-coherent and visually appealing results than the state-of-the-art methods on inpainting high resolution videos.