Demo: Fast computation of seamless video loops

Version 1.0 — 2015-10-25.

Version 1.1 — 2015-11-12. Fixed bug when rendering on Intel HD Graphics.

This package demonstrates the technology in the research paper:

Fast computation of seamless video loops.
Jing Liao, Mark Finch, Hugues Hoppe.
ACM Trans. Graphics (SIGGRAPH ASIA), 34(6), 2015.

Given a short (~5 second) input video, we automatically compute a new video that (1) can loop endlessly and (2) minimizes spatial and temporal artifacts. Please refer to the project web page and additional results.

Requirements

This demo code runs under 64-bit Windows 8 or Windows 10. Creating video loops requires ≥ 4 GB of memory (depending on video resolution and length). The package should contain all the necessary redistributable *.dll files. (If there are errors regarding missing dll files and Microsoft Visual Studio 2015 is not installed, it may be necessary to install its runtime components.)

Download

The demo can be found here.

Package contents

The demo consists of an executable program and several batch scripts:

Loop creation

Video reading and writing is performed using Windows Media Foundation. Supported video formats include *.mp4, *.wmv, and *.avi files.

By default, the viewer selects up to 5 seconds of the input video as source content. If the video is 7 seconds or longer, then the section from second 2 to second 7 is used, otherwise the last 5 seconds of the video is used. The automatic selection can be overridden by setting trim IN and OUT frames using the < and > keys in VideoViewer.

Video stabilization

The loop creation program assumes that input videos have no camera motion. This is best achieved by shooting videos with a camera on a tripod. If the video is hand-held, visible camera motion can lead to artifacts resulting in a poorer quality looped video. Thus, for hand-held videos we recommend running a software stabilizer or alignment process first. Here are some resources for stabilizing videos: