Cycle-Consistent Counterfactuals by Latent Transformations (C3LT)

PyTorch implementation for C3LT, a novel counterfactual visual explanation method published at CVPR 2022.

• Saeed Khorram, Li Fuxin. "Cycle-Consistent Counterfactuals by Latent Transformations (C3LT)", CVPR 2022.

Dependencies

First install and activate a python3.6 virtual environment:

$ python3.6 -m venv env
$ source env/bin/activate

You can update the pip and install the dependencies using:

(env) $ pip install --upgrade pip
(env) $ pip install -r req.txt

Quick Start

For instance, to train CF latent transformations for classes 4 and 9 from the mnist dataset, one can simply run:

(env) $ python main.py --dataset mnist --cls_1 4 --cls_2 9

The hyperparameters for training can be directly passed as arguments when running the main.py. For the full list of arguments, please see args.py.

Citation

If you use the implementation in your research, please consider citing our paper:

@inproceedings{khorram2022cycle,
  title={Cycle-Consistent Counterfactuals by Latent Transformations},
  author={Khorram, Saeed and Fuxin, Li},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}