Skip to content

Latest commit

 

History

History
29 lines (23 loc) · 1.36 KB

README.md

File metadata and controls

29 lines (23 loc) · 1.36 KB
Raw

PICO - Pragmatic Image Compression

PICO is a lossy image compression algorithm that adapts compression to user behavior, optimizing reconstructions to be useful for downstream tasks instead of preserving visual appearance.

Usage

  1. Clone pico into your home directory ~
  2. Setup an Anaconda virtual environment with conda create -n picoenv python=3.6
  3. Install dependencies with pip install -r requirements-{a,b}.txt (a for mnist and carracing, b for celeba and lcars)
  4. Replace NVAE/model.py with deps/NVAE/model.py
  5. Replace gym/envs/box2d/car_{dynamics,racing}.py with deps/box2d/car_{dynamics,racing}.py
  6. Replace gym/envs/classic_control/rendering.py with deps/classic_control/rendering.py
  7. Replace stylegan2/projector.py with deps/stylegan2/projector.py
  8. Install the pico package with python setup.py install
  9. Jupyter notebooks in pico/notebooks provide an entry-point to the code base

Citation

If you find this software useful in your work, we kindly request that you cite the following paper:

@article{pico2021,
  title={Pragmatic Image Compression for Human-in-the-Loop Decision-Making},
  author={Reddy, Siddharth and Levine, Sergey and Dragan, Anca D.},
  journal={arXiv preprint arXiv:2108.04219},
  year={2021}
}