https://github.com/materight/understandable-protopnet
A study on the interpretability of the concepts learned by Prototypical Part Networks (ProtoPNets) on the CUB200-2011 and CelebAMask datasets.
https://github.com/materight/understandable-protopnet
celeba cub200 part-prototype-networks prototypical-networks xai
Last synced: 3 months ago
JSON representation
A study on the interpretability of the concepts learned by Prototypical Part Networks (ProtoPNets) on the CUB200-2011 and CelebAMask datasets.
- Host: GitHub
- URL: https://github.com/materight/understandable-protopnet
- Owner: materight
- Created: 2021-12-17T13:03:32.000Z (over 3 years ago)
- Default Branch: main
- Last Pushed: 2023-03-04T19:59:55.000Z (about 2 years ago)
- Last Synced: 2025-01-04T05:52:26.214Z (5 months ago)
- Topics: celeba, cub200, part-prototype-networks, prototypical-networks, xai
- Language: Python
- Homepage:
- Size: 1.65 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# ProtoPNet: Are Learned Concepts Understandable?
A study on the interpretability of the concepts learned by [Prototypical Part Networks](https://arxiv.org/abs/1806.10574) (ProtoPNets).
This work exploits the part locations annotations available for two different datasets to provide an objective evalution of the prototypes. An additional *diversity regularization* is also introduced to produce more diverse concepts.
More details on the implementation can be found in the [report](report.pdf).
California Gull class
580![]()
![]()
583![]()
![]()
585![]()
![]()
587![]()
![]()
588
Female class
2![]()
![]()
3![]()
![]()
4![]()
![]()
6![]()
![]()
7
## Get started
- Clone the repository and install the required dependencies:
```shell
git clone https://github.com/materight/explainable-ProtoPNet.git
cd explainable-ProtoPNet
pip install -r requirements.txt
```
- Download and prepare the data, either for the [Caltech-UCSD Birds-200](http://www.vision.caltech.edu/datasets/cub_200_2011/) or the [CelebAMask HQ](http://mmlab.ie.cuhk.edu.hk/projects/CelebA/CelebAMask_HQ.html) datasets:
```shell
python prepare_data.py cub200
python prepare_data.py celeb_a
```
## Train a model
To train a new model on a dataset, run:
```shell
python train.py --dataset [data_path] --exp_name [experiment_name]
```
Additional options can be specified (run the script with `--help` to see the available ones).
After training, the learned prototypes can be further pruned:
```shell
python prune_prototypes.py --dataset [data_path] --model [model_path]
```
## Evaluate learned prototypes
To evaluate a trained model and the learned prototypes, run:
```shell
python evaluate.py --model [model_path] {global|local|alignment} --dataset [data_path]
```
- `global`: retrieve for each prototype the most activated patches in the whole dataset.
- `local`: evaluate the model on a subset of samples and generate visualizations for the activated prototypes for each class.
- `alignment`: generate plots for the alignment matrix of each class.
## Acknowledgments
This implementation is based on the original [ProtoPNet](https://github.com/cfchen-duke/ProtoPNet) repository.