I'd like to see this in-person. looks pretty cool to me!

Does anyone know how they display 45 different angles of an 3d object without using a ton of computing power? I'm imagining you'd need a beast of a computer to use this display at a reasonable refresh rate without some really clever optimization.

I'm pretty sure it's a lenticular display, so if it's running a standard 8K panel (~33 megapixel), your effective resolution is under 1 megapixel per angle.

Computationally, generating 33 megapixels spread across 45 views is very similar to generating 33 megapixels in one view, so if your rendering pipeline can do standard 8K, you should be fine.

You can get a lot done with re-projection and temporal anti-aliasing. At the most abstract level you can consider it an encoding problem: given an image, what's the marginal entropy of a slightly deviating image? There's very little that actually needs to be uniquely computed.

Edit: If you're willing to give up strict coherence between views you might be able to do even better.

Please someone code a neural net to add 3D paint depth to 2D-photo of real paintings :)

A training set for this could be created from VR paintings. We have the full spatial painting and we can create as many 2D views as we want from it.