--- dataset_info: - config_name: margin025 features: - name: file dtype: image - name: age dtype: string - name: gender dtype: string - name: race dtype: string - name: service_test dtype: bool splits: - name: train num_bytes: 513550350.352 num_examples: 86744 - name: val num_bytes: 64534471.096 num_examples: 10954 download_size: 563297165 dataset_size: 578084821.448 - config_name: margin125 features: - name: file dtype: image - name: age dtype: string - name: gender dtype: string - name: race dtype: string - name: service_test dtype: bool splits: - name: train num_bytes: 1860789457.104 num_examples: 86744 - name: val num_bytes: 236793098.794 num_examples: 10954 download_size: 2104353129 dataset_size: 2097582555.898 configs: - config_name: margin025 data_files: - split: train path: margin025/train-* - split: val path: margin025/val-* - config_name: margin125 data_files: - split: train path: margin125/train-* - split: val path: margin125/val-* license: cc-by-4.0 --- # Dataset Card for FairFace ## Table of Contents - [Dataset Card Creation Guide](#dataset-card-creation-guide) - [Table of Contents](#table-of-contents) - [Dataset Description](#dataset-description) - [Dataset Summary](#dataset-summary) - [Dataset Structure](#dataset-structure) - [Data Instances](#data-instances) - [Data Fields](#data-fields) - [Additional Information](#additional-information) - [Licensing Information](#licensing-information) - [Citation Information](#citation-information) ## Dataset Description - **Repository:** [https://github.com/joojs/fairface](https://github.com/joojs/fairface) - **Paper:** [FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation](https://openaccess.thecvf.com/content/WACV2021/html/Karkkainen_FairFace_Face_Attribute_Dataset_for_Balanced_Race_Gender_and_Age_WACV_2021_paper.html) ### Dataset Summary A dataset of human faces annotated with discrete categories for the photographed person's age, sex, and race. Please consider prioritizing [a previously created Hugging Face dataset repository for Fair Face](https://huggingface.co/datasets/HuggingFaceM4/FairFace) as this new dataset repository was only made for downloading issues that may already be resolved. For complete details on the dataset's construction and intended uses, please refer to the dataset's official repository or paper. ## Dataset Structure ### Data Instances Each instance contains an image and discrete categories for age, gender, and race. ``` { 'file': , 'age': '50-59', 'gender': 'Male', 'race': 'East Asian', 'service_test': True } ``` ### Data Fields - `file`: an image of a human face with either padding = 0.25 or padding = 1.25 depending on the dataset config - `age`: a category describing the age of the person in the image limited to *0-2*, *3-9*, *10-19*, *20-29*, *30-39*, *40-49*, *50-59*, *60-69*, and *more than 70* - `gender`: a category describing the sex of the person in the image limited to *Male* and *Female* - `race`: a category describing the race of the person in the image limited to *East Asian*, *Indian*, *Black*, *White*, *Middle Eastern*, *Latino_Hispanic*, and *Southeast Asian* - `service_test`: please refer to this [issue](https://github.com/dchen236/FairFace/issues/20) from the dataset's official repository ## Additional Information ### Licensing Information According to the official repository, FairFace is licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). ### Citation Information ``` @InProceedings{Karkkainen_2021_WACV, author = {Karkkainen, Kimmo and Joo, Jungseock}, title = {FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation}, booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, month = {January}, year = {2021}, pages = {1548-1558} } ```