Disney Research has posted an extremely cool tech demo of DuctTake, a new system for seamlessly compositing multiple takes of a scene into a single shot with minimal manual clean-up.
To control the process, a user simply draws (very roughly) onto the image to identify which part must be retained from each take.
DuctTake automatically combines the takes by “finding optimal spatiotemporal seams using motion-compensated 3D graph cuts through the video volume”.
The demo then shows the system calculating a grid of optical flow vectors to match video streams seamlessly.
Solves common compositing problems
Use cases shown in the demo include combining the best parts of two takes, removing unwanted objects or people from a shot – or simply goofing around.
Towards the end, there’s an interesting comparison with Nuke’s Furnace Align plugin (the researchers claim their solution has more temporal stability), reinforcing just how easy DuctTake makes a hard job look.
Read more about DuctTake on the Disney Research website
(Includes the full research paper)