search

zhengqili / Neural-Scene-Flow-Fields

PyTorch implementation of paper "Neural Scene Flow Fields for Space-Time View Synthesis of Dynamic Scenes"
MIT License
707 stars 94 forks source link

MiDaS depth prediction -- inverse depth? #27

Open b0ku1 opened 2 years ago

b0ku1 commented 2 years ago

Hi,

I just want to quickly inquire about Midas depth prediction. The original midas approach seems to predict disparity or inverse depth, rather than euclidean depth: https://github.com/isl-org/MiDaS/issues/42.

However, as far as I can tell, the rendered depth map is based on z-values distance, not inverse depth. Thus, is your MiDaS model pretrained on depth? Thanks.

zhengqili commented 2 years ago

We use pretrained Midas network. We optimize NSFF in ndc space as original NeRF, and the depth in NDC space is equal to negative disparity in Euclidean space up to scale and shift as derived in https://www.cs.cornell.edu/~zl548/NSFF/NSFF_supp.pdf