Each scene includes the following files:

1. subset_sphere_0_data_clean.json: JSON file that stores all the information for trainning/evaluation. In particular, it includes
 1) img_name: name of image 
 2) rel_trans: translation relative to the reference viewpoint, the first image is always the reference viewpoint we choose (i.e. zero translation)
 3) K: intrinsic matrix 
 4) R_cg, t_cg: 3X3 rotation and 3x1 translation from global coordinate system to local camera coordindate system.
 5) near_plane_depth: near plane depth computed from SfM points

2. images folder: undistorted images of each scene 

3. images_rotate: undistorted images rotated to upright orientation (based on estimated roll angle from SfM).

4. semantics_rotate: semantic segmentation rotated to upright orientation

5. angles: camera roll angles estimated from SfM.
 
6. sparse: sparse SfM model in COLMAP format

7. warp_imgs: warped rectified images at reference viewpoint used for apperance encoding.

8. viz_sphere_64_clean_0: average Plane Sweep Volume (PSV) computed from Internet photos, which are used in stage one of our system.