See oculomotor for the hackathon architecture.

• data
  • data extractor for beta VAE training
• oculomotor
  • main execution files

cd data
python oddoneoutgen.py --episode=<int> --length=<int> --scene=<int>

• create egocentric datasets
  • shape: (episode*length, height=128, width=128, channel=3)
  • save: data/images/OddOneOut/egocentric_images%Y%m%d.npy
• create allocentric datasets
  • shape: (scene, height=128, width=128, channel=3)
  • save: data/images/OddOneOut/allocentric_images%Y%m%d.npy