Real-Time Selfie Video Stabilization

TL;DR

This paper introduces a real-time, automatic selfie video stabilization method using a 1D convolutional network to achieve high-quality stabilization at 26 fps, with user-controlled focus and a large training dataset.

Contribution

The paper presents a novel real-time selfie video stabilization approach with a specialized network, a new dataset, and a grid approximation for fast warping, outperforming previous methods.

Findings

Runs at 26 fps with real-time performance

Produces better visual and quantitative stabilization results

Achieves comparable quality to offline methods with much faster speed

Abstract

We propose a novel real-time selfie video stabilization method. Our method is completely automatic and runs at 26 fps. We use a 1D linear convolutional network to directly infer the rigid moving least squares warping which implicitly balances between the global rigidity and local flexibility. Our network structure is specifically designed to stabilize the background and foreground at the same time, while providing optional control of stabilization focus (relative importance of foreground vs. background) to the users. To train our network, we collect a selfie video dataset with 1005 videos, which is significantly larger than previous selfie video datasets. We also propose a grid approximation method to the rigid moving least squares warping that enables the real-time frame warping. Our method is fully automatic and produces visually and quantitatively better results than previous real-time general video stabilization methods. Compared to previous offline selfie video methods, our approach produces comparable quality with a speed improvement of orders of magnitude.