This is the official repository for the Interpretability Dataset "Banapple". This dataset is associated with the paper "E pluribus unum interpretable convolutional neural networks". The code for the paper can be found here.
For the purposes of the study presented in "E pluribus unum interpretable convolutional neural networks", a novel dataset was constructed, named Banapple. The dataset consists of images of bananas and apples. It was created by collecting images, under the Creative Commons license, from Flickr.
The images illustrate bananas and apples with variations regarding the color, placement, size, and background. The motivation for the construction of this dataset stems from studies in cognitive science, where human perception is investigated using examples with discrete properties of bananas and apples. The experiments performed aim to demonstrate that EPU-CNN is capable of capturing the discriminative characteristics of bananas and apples by the perceptual features it incorporates, i.e., apples have a circular shape and usually red color, whereas bananas have a bow-like shape and usually a yellow color.
In addition, samples that deviate from the average appearance of these objects can provide insights regarding the reliability of the interpretation of the model. The dataset consists of images of bananas and apples.
Banapple consists of 2,313 images of bananas and apples. The images are divided into three partitions for training, validation and testing.
- Training: 1,666 images
- Validation: 417 images
- Testing: 230 images
The images are in JPEG format and have various sizes. The filenames of the images are in the
format {class}{imageIndex}.jpg},
where class is either apple or banana, and imageIndex is a number between 1 and 2313, e.g., apple458.jpg.
Some samples from the dataset are shown below.
| Class: apple | Class: banana |
|---|---|
 |
 |
 |
 |
 |
 |
You can download the dataset from here.
If you find this work useful, please cite our paper:
@article{dimas2023pluribus,
title = {E pluribus unum interpretable convolutional neural networks},
author = {Dimas, George and Cholopoulou, Eirini and Iakovidis, Dimitris K},
journal = {Scientific Reports},
volume = {13},
number = {1},
pages = {11421},
year = {2023},
publisher = {Nature Publishing Group UK London}
}
- Add a download script
- Add a dataset management Class (e.g.
BanappleDataset) - Replace the .arxiv reference with the Scientific Reports reference