Official Tensorflow implementation of ICLR 2020 paper: Explanation By Progressive Exaggeration.
Training graphs and qualitative results
./notebooks/Process_Classifier_Output.ipynb
$ pip install -r requirements.txt- Download the CelebA dataset and create train and test fold for the classifier.
CelebA dataset is downloaded and saved at ./data/CelebA. IPython notebook creates text files with file names and labels and save them at ./data/CelebA/. These text files are used as input data to train the classifier.
./notebooks/Data_Processing.ipynb
- Train a classifier. Skip this step if you have a pretrained classifier. The output of the classifier is saved at:
$log_dir$ /$name$.
2.a. To train a multi-label classifier on all 40 attributes
python train_classifier.py --config 'configs/celebA_DenseNet_Classifier.yaml'
2.b. To train a binary classifier on 1 attribute
python train_classifier.py --config 'configs/celebA_Smile_DenseNet_Classifier.yaml'
- Process the output of the classifier and create input for Explanation model by discretizing the posterior probability.
The input data for the Explanation model is saved at:
$log_dir$ /$name$/explainer_input/
./notebooks/Process_Classifier_Output.ipynb
- Trainer explainer model. The output is saved at:
$log_dir$ /$name$.
python train_explainer.py --config 'configs/celebA_Young_Explainer.yaml'
- Explore the trained Explanation model and see qualitative results.
./notebooks/Test_Explainer.ipynb
- Save results of the trained Explanation model for quantitative experiments.
python test_explainer.py --config 'configs/celebA_Young_Explainer.yaml'
- Use the saved results to perform experiments as shown in paper
./notebooks/Experiment_CelebA.ipynb
@inproceedings{
Singla2020Explanation,
title={Explanation by Progressive Exaggeration},
author={Sumedha Singla and Brian Pollack and Junxiang Chen and Kayhan Batmanghelich},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=H1xFWgrFPS}
}