Training on objects

[lalalune]

Hello,

I am curious if the implementation would be capable of novel view synthesis of objects if trained on multiview object data, as opposed to scene data? Do you see any potential blockers?

Thanks!