Summit Data

Summit is an interactive system that scalably and systematically summarizes and visualizes what features a deep learning model has learned and how those features interact to make predictions. This repository contains the data used in the Summit visualization.

For the main Summit repo, go to https://github.com/fredhohman/summit.

Data

All data in the context of InceptionV1 trained on ImageNet.

• data/imagenet.json: metadata for each class including aggregated activations
• data/feature-vis/: feature visualizations images
  • data/feature-vis/channel/: channel feature visualizations
  • data/feature-vis/diversity-0/: diversity feature visualizations (1/4)
  • data/feature-vis/diversity-1/: diversity feature visualizations (2/4)
  • data/feature-vis/diversity-2/: diversity feature visualizations (3/4)
  • data/feature-vis/diversity-3/: diversity feature visualizations (4/4)
  • data/feature-vis/dataset-p/: positive dataset examples for each channel
• data/ag/: attribution graphs for each class

Live Demo

For a live demo, visit: fredhohman.com/summit

Download

Download or clone this repository (warning: this repository is big (~1GB) ):

git clone https://github.com/fredhohman/summit-data.git

Resources

We use feature visualizations from Feature Visualization, Distill 2017, and Lucid.

License

MIT License. See LICENSE.md.

Citation

@article{hohman2020summit,
  title={Summit: Scaling Deep Learning Interpretability by Visualizing Activation and Attribution Summarizations},
  author={Hohman, Fred and Park, Haekyu and Robinson, Caleb and Chau, Duen Horng},
  journal={IEEE Transactions on Visualization and Computer Graphics (TVCG)},
  year={2020},
  publisher={IEEE},
  url={https://fredhohman.com/summit/}
}

Contact

For questions or support open an issue or contact Fred Hohman.