NVlabs / imaginaire

NVIDIA's Deep Imagination Team's PyTorch Library
Other
4.01k stars 450 forks source link

wc_vid2vid with Synthetic data from a Game Engine ? #68

Closed DanAmador closed 3 years ago

DanAmador commented 3 years ago

Good day,

I'm trying to generate a dataset for the wc_vid2vid model.

I've managed to get the semantic segmentation images, camera extrinsics (world position and rotation) and rgb-d images.

What I still don't fully understand is: what exactly do the unprojection values mean? In the unprojection script theres a "idx_3d" variable, is this just an id to that particular point in space?

Can I use my collected data to generate the unprojection files without having to use OpenSFM and the unprojection script?