Data and Code Access

We have released all Python code needed to reproduce our experiments as well as to download and process our dataset:

Methods Code on GitHub Dataset Code on GitHub

The raw dataset is available in ZIP format:

Download NovelCraft (1.5GB) Download NovelCraft+ (4.5GB)

We ask that you approach this dataset responsibly:

  • Pursue applications with positive benefit to society
  • Please cite our paper in any published work using the dataset.

Extra Training Set: NovelCraft+

As of August 2022, we released an extra supplement called NovelCraft+, prounced as (“novel craft plus”).

This dataset contains 130,533 additional frames of standard environment training data, significantly increasing the dataset size.


Our dataset is available under a CC-BY 4.0.

Storage requirements

Original Dataset

The original dataset is available as a single large ZIP file, which will require ~1.5 GB.

After uncompressing, the NovelCraft dataset will require about ~4.5 GB.

NovelCraft+ (aka NovelCraftPlus)

Our large NovelCraft+ dataset is available as a ZIP file which will require ~4.5 GB.

After uncompressing, NovelCraft+ requires an additional ~12.2 GB.