Data structure

• JSON file indicating input and test views for every scene (e.g. Track1_split_test.json for the Test split, Track 1):
  • keys in the JSON corresponding to scene name (and directory name),
  • for every key, a dictonary containining the following fields
    • input_views indicating indices of views used as an input (only those are provided)
    • test_views indicating indices used for testing required in the submission
    • dataset indicating whether data comes from DTU or Synthetic SpaRe part (remember that index i corresponds to camera name (i + 1) - see Evaluation section).
• Calibration data for DTU and Synthetic datasets provided as projection matrices, compatible with common data loading codes (e.g.as in RegNeRF),
• Directories corresponding to scene name, e.g. scan63, or Amplifier.
• Within directories, images corresponding to various capture positions in various lighting conditions, named as following:
  • for DTU files: rect_{idx+1:03d}_{lighting}_r5000.png
  • for SpaRe files: cam_{idx+1:03d}_{lighting}_0000.png

  where idx corresponds to view index listed in the JSON file, and lighting corresponds to lighting conditions from 0 to 6, where 3 indicates all light sources on, and 3 is the setup used for evaluation in this competition.