search

EPFL-VILAB / omnidata

A Scalable Pipeline for Making Steerable Multi-Task Mid-Level Vision Datasets from 3D Scans [ICCV 2021]
Other
395 stars 49 forks source link

How to use the annotator with only mesh and RGB images? #16

Closed ewrfcas closed 1 year ago

ewrfcas commented 2 years ago

Thanks for your good works! The paper said that the annotator can be extended to Tanks and Temples. I would like to know how to use the annotator to render depths for Tanks and Temples with only mesh(no texture) and RGB images(no camera poses)? And what does 'aligned' in 'aligned RGB images' mean in this paper?

alexsax commented 1 year ago

Hi @ewrfcas

'Aligned' in this case means that we do have camera pose for the RGB images. For T+T the camera poses are available.

Using the mesh for geometry and RGB + camera poses for images is how we generated the Taskonomy and 2D3DS datasets. You can look at the demo for an example, though the demo uses textured meshes. The only difference is that you want to use the script create_rgb_images.py instead of create_rgb_images_textured.py. As for how to do this, see the entrypoint run_demo.sh