applicability issue to realworld data

[admiral-chen]

Hello,

Awesome work! I have two questions regarding the applicability of Matrix-3D. First, as the training data Matrix-Pano is curated from simulation data, I wonder whether Matrix-3D is applicable to real world images such as DLV3D/RealEstate10k? 2) Can the optimization-based reconstruction branch generate rendered perspective videos, just like the feed-forward reconstruction branch can do?

Thanks a lot!