Datasets for the evaluation of explainability methods for computer vision models
doi:10.4121/c5831e59-8e02-444d-b753-99371b93724e.v1
The doi above is for this specific version of this dataset, which is currently the latest. Newer versions may be published in the future.
For a link that will always point to the latest version, please use
doi: 10.4121/c5831e59-8e02-444d-b753-99371b93724e
doi: 10.4121/c5831e59-8e02-444d-b753-99371b93724e
Datacite citation style:
Balayn, Agathe (2023): Datasets for the evaluation of explainability methods for computer vision models. Version 1. 4TU.ResearchData. dataset. https://doi.org/10.4121/c5831e59-8e02-444d-b753-99371b93724e.v1
Other citation styles (APA, Harvard, MLA, Vancouver, Chicago, IEEE) available at Datacite
Dataset
usage stats
213
views
105
downloads
licence
CC BY-NC 4.0
Set of datasets of images, their ground truth, their saliency map for one model, and the manual annotations of the saliency maps with semantic concepts of various granularities. The images in each set were selected in order to inject class-specific biases in the dataset.
- One set consists of two classes sampled from ImageNet, representing ambulances and vans.
- One set consists of two classes sampled from PA-100K, representing human males and females.
- 4 sets consists of three classes extracted from ImageNet, lobsters, sharks, and trouts, sampled in different ways to include various biases.
history
- 2023-03-30 first online, published, posted
publisher
4TU.ResearchData
format
zipped file, containing images (.jpg, .png), and annotations (.csv)
associated peer-reviewed publication
What do You Mean? Interpreting Image Classification with Crowdsourced Concept Extraction and Analysis
references
organizations
TU Delft, Faculty of Electrical Engineering, Mathematics and Computer Science, Department of Software Technology
DATA
files (2)
- 1,614 bytesMD5:
e73a533538d8327ec787eb6fc2afee8a
README.md - 298,694,270 bytesMD5:
5e94ca074aa3a27eb28929bef7cb8836
Archive.zip -
download all files (zip)
298,695,884 bytes unzipped