Skip to content

Latest commit

 

History

History
24 lines (23 loc) · 1.82 KB

training_data.md

File metadata and controls

24 lines (23 loc) · 1.82 KB

Training Data

We provide a few suggestions on how to setup training data for GRIT. We focus here on how to get data and annotations from datasets used in GRIT that you are allowed to train on.

Surface Normals

Many of the surface normals in GRIT have been created from some other type of 3D data (e.g. depth, point clouds). Therefore, to get surface normals for the images in the GRIT dataset, you either need to generate them yourself or download the normals from someone else who has already done this.

ScanNet

For ScanNet images, we use surface normals created by the FrameNet folks. Use the provided download script: