Metadata-Version: 2.1
Name: sc2image
Version: 1.0.4
Summary: The StarCraft Image dataset
Home-page: https://starcraftdata.davidinouye.com/
Author: StarCraftImage team
Author-email: dinouye@purdue.edu
License: MIT
Platform: UNKNOWN
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Requires-Python: >=3.7
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: torch (>=1.7.0)
Requires-Dist: numpy (>=1.19.1)
Requires-Dist: tqdm (>=4.53.0)
Requires-Dist: torchvision (>=0.8.2)
Requires-Dist: outdated (>=0.2.0)
Requires-Dist: pandas (>=1.1.0)
Requires-Dist: pillow (>=7.2.0)
Requires-Dist: tqdm (>=4.6.0)

# StarCraftImage Dataset
<p align='center'>
  <img width='max%' src='https://github.com/inouye-lab/starcraftimage/blob/main/figures/dataset-overview-figure.png' />
</p>

--------------------------------------------------------------------------------

[![PyPI](https://img.shields.io/pypi/v/sc2image)](https://pypi.org/project/sc2image/)
[![License](https://img.shields.io/github/license/inouye-lab/starcraftimage)](https://github.com/inouye-lab/starcraftimage/blob/main/LICENSE)


Welcome! This is the repository for the StarCraftImage dataset from the paper: [StarCraftImage: A Dataset For Prototyping Spatial Reasoning Methods For Multi-Agent Environments](https://openaccess.thecvf.com/content/CVPR2023/html/Kulinski_StarCraftImage_A_Dataset_for_Prototyping_Spatial_Reasoning_Methods_for_Multi-Agent_CVPR_2023_paper.html).
This README has basic installation and quickstart usage but our project webpage has more documentation: [StarCraftImage project webpage](https://starcraftdata.davidinouye.com/). 

If you use this dataset, please cite the following paper:
```
@inproceedings{kulinski2023starcraftimage,
  title={StarCraftImage: A Dataset for Prototyping Spatial Reasoning Methods for Multi-Agent Environments},
  author={Kulinski, Sean and Waytowich, Nicholas R and Hare, James Z and Inouye, David I},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={22004--22013},
  year={2023}
}
```

## Installation

We recommend using the `pip` package manager to install `sc2image`.

```bash
pip install sc2image
```

Note, this does not include the `dataset-demos` folder which contains jupyter notebooks which show example uses of the dataset.
To use them, you should install from source via:
    
```bash
git clone git@github.com:inouye-lab/starcraftimage.git
cd starcraftimage
pip install -e .
```

## Quickstart

:warning: **For more detailed documentation and example uses, visit the [dataset docs](https://starcraftdata.davidinouye.com/docs) for using `sc2image`.**

There are three main StarCraftII datasets. 
Each dataset includes images summarize a 10 second window (255 frames) of a StarCraftII replay.

1. `StarCraftImage`: This is the main 3.6 million sample dataset which includes multiple image formats:`'sparse-hyperspectral'`, `'dense-hyperspectral'`, `'bag-of-units'`, `'bag-of-units-first'`, and contains all unit positioning information throughout the window.
This dataset can be used via the following:

    ```py
    from sc2image.dataset import StarCraftImage
    sc2image = StarCraftImage(root="data", download=True)
    ```

    This will download the StarCraftImage dataset to the `data` directory (if it does not already exist there).
As this dataset has over 3.6 million samples, this might take a while to download. However, you can use the standalone StarCraftCIFAR10 and StarCraftMNIST versions below.


2. `StarCraftCIFAR10`: This is a simplified version of the `StarCraftImage` dataset which exactly matches the setup of the CIFAR10 dataset.
All images have been condensed into a three channel (RGB) image where the Red channel corresponds to Player 2 units, Green correspond to neutral units, and Blue to Player 1 units.
The 10 classes equate to: `(map_name, did_window_happen_in_first_half_of_replay)`.
The dataset can be loaded via:
        
    ```py    
    from sc2image.dataset import StarCraftCIFAR10
    sc2image_cifar10 = StarCraftCIFAR10(root="data", download=True)
    ```

 3. `StarCraftMNIST`: This is a further simplified version of the `StarCraftImage` dataset which exactly matches the setup of the MNIST dataset. 
 The grayscale images show to the seen last seen timestamps for units each pixel location, and the 10 classes match that of `StarCraftCIFAR10`.
 The dataset can be loaded via:

    ```py
    from sc2image.dataset import StarCraftMNIST
    sc2image_mnist = StarCraftMNIST(root="data", download=True)
    ```
    
## Example uses
Please see the `starcraftimage-quickstart` jupyter notebook in the `dataset-demos` folder to see details on using this dataset!

## Bug reports

If you run into any issues, please feel free to open an issue in this repository or email us via the corresponding author email in the [main paper](https://openaccess.thecvf.com/content/CVPR2023/html/Kulinski_StarCraftImage_A_Dataset_for_Prototyping_Spatial_Reasoning_Methods_for_Multi-Agent_CVPR_2023_paper.html).

Cheers!


