search

autonomousvision / giraffe

This repository contains the code for the CVPR 2021 paper "GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields"
https://m-niemeyer.github.io/project-pages/giraffe/index.html
MIT License
1.24k stars 160 forks source link

[Question about Equation7][transform points from object to scene space] #9

Closed tomguluson92 closed 3 years ago

tomguluson92 commented 3 years ago

Dear author, I am really inspired by your brilliant work! However, I found a simple point struck me, As Eq6 says: k(x) "it transform points from object to scene space" while I found its usage is the k^-1(x) in Eq7. Would you be kind to explain this? Thank you very much!

image

MrTornado24 commented 3 years ago

k(x) means the transformation from object to scene, but in their code, they actually sample x from world coordinates, and they need to use k-1(x) to transform x from scene to object space

m-niemeyer commented 3 years ago

Here @tomguluson92 , thanks for your question. In a forward pass, we sample random transformations. They are then used here to transform points from 3D scene space to the canonical object space. You can find the transformation function here. As @MrTornado24 correctly pointed out, we directly use k^-1 because as pointed out in Eq 7, this is what we need to evaluate.

Hope this helps a little!