The Spinning Statue.

Give me two images and I'll give you the world! 3D world to be exact...
But do you really want a 3D model?
Accurate 3D model is hard to obtain and generating a new view from a 3D model depends on the complexity of the scene.

We can do it without the 3D model!!!

It has been established that three matching measurements accross three views satisfy linear constraints. These constraints, called the trilinear tensor are accurate, robust and enable direct novel view synthesis of a third image without explicit 3D representation.
Hard to Believe ???

Just Watch Us!

Here is a pair of uncalibrated images taken with an indycam.
We have no idea what are the internal or external parameters of the cameras

Still, we have generated the following:

Video Fly Around.

[Back] Back to Novel Views
[Back] Back to more demos