search

TRI-ML / packnet-sfm

TRI-ML Monocular Depth Estimation Repository
https://tri-ml.github.io/packnet-sfm/
MIT License
1.21k stars 241 forks source link

How to calculate our own ray surface template? #249

Closed 942411526 closed 1 year ago

942411526 commented 1 year ago

we want to calculate the ray surface template from our dataset.

VitorGuizilini-TRI commented 1 year ago

Do you have a camera model you want to use as the template, or are you creating your own from scratch?

942411526 commented 1 year ago

Do you have a camera model you want to use as the template, or are you creating your own from scratch?

we have some data on unknown camera parameters(such as other datasets and video from YouTube ) and some data captured by our camera realsense D435i

VitorGuizilini-TRI commented 1 year ago

A ray surface template is a HxWx3 tensor with viewing rays for each pixel on your image, so if you have any sort of camera model you can create it by unprojecting points to 3D space following that model, and normalizing them to unit length. After that, our network learns an offset that goes from that template to the "true" camera model.

942411526 commented 1 year ago

A ray surface template is a HxWx3 tensor with viewing rays for each pixel on your image, so if you have any sort of camera model you can create it by unprojecting points to 3D space following that model, and normalizing them to unit length. After that, our network learns an offset that goes from that template to the "true" camera model.

Thanks for your response very much. I will do that.

VitorGuizilini-TRI commented 1 year ago

Great, let me know if it works, I'm closing this for now