Information regarding the datasets

#1

There have been some questions regarding the test and training datasets for the challenge. Some of the information is summarized below. We request all the participants to ask the relevant questions at this forum.

  1. All the datasets have the same factors and number of images. Check out the arXiv preprint of On the Transfer of Inductive Bias from Simulation to the Real World: a New Disentanglement Dataset for more information.
  2. To avoid any overfitting to the datasets, unseen/hidden factors can be used for fair evaluations.
  3. The real-world dataset is captured in a controlled environment and the conditions are kept as consistent as possible.
  4. The dims of the images are the same i.e. 64x64.