search

autonomousvision / differentiable_volumetric_rendering

This repository contains the code for the CVPR 2020 paper "Differentiable Volumetric Rendering: Learning Implicit 3D Representations without 3D Supervision"
http://www.cvlibs.net/publications/Niemeyer2020CVPR.pdf
MIT License
794 stars 90 forks source link

Output camera projection matrix? #53

Closed Kulbear closed 3 years ago

Kulbear commented 3 years ago

Hello,

Thanks for your great work on this problem! I am wondering whether it is also possible to get the projection matrix that can map the output mesh back to the 2D image?

Thanks!

m-niemeyer commented 3 years ago

Hi @Kulbear , thanks for your interest in our project!

We use the transform_to_world function to project pixels with a depth value to the world space. There, you can also find the inverse transform, transform_to_camera_space which is the function which maps 3D world points to the camera space. To get out the final pixel values, you have to divide the first two coordinates by the third which is the depth in camera space. To give an example, we are using the function here to get the depth value in camera space.

Good luck with your research!

wangsen1312 commented 3 years ago

@m-niemeyer Thank you for your great work! Can you tell me which configuration can generate the predict camera_mat/world_mat/Scale_mat along with the meshes?