Interpolating 3D Stereo Pairs
Presenting 3D stereoscopic images is a challenge. Personally I usually view them in cross-eyed free viewing mode (like so), but that takes training (and sometimes eye strain). The most accessible method of displaying them to untrained viewers is the “wiggly gif”. This conveys depth pretty well, but can be jarring:
For a while now I’ve had my eyes on Simon Eugster’s excellent slowmoVideo software. It uses Optical Flow analysis to determine motion between frames, and then synthesizes intermediate frames to allow slowing down regular video footage without making it jerky. Conveniently, the same technique works really well at generating smooth transitions between stereo pairs.
Until a few days ago it required an nvidia graphics card (which I don’t have), but the recent 0.2.6 release can make use of the OpenCV software library instead - so I was thrilled to finally be able to try generating a smoothly interpolated version of some stereo pairs. Here’s the earlier stereo pair with just a few intermediate frames synthesized:
It’s only a few extra frames, but the effect is quite amazing. It sadly didn’t work on all of the stereo pairs I tried - if there is too much motion between frames it can get confused and produce weird artifacts, but when it works it’s pretty damn magical.
It’s a bit of a manual process at the moment, but I’m hoping to automate the laborious bits and integrate it into my stereoscoper command-line tool to allow generating these with minimal effort.