README.md

metadata

dataset_info:
  features:
    - name: Image file
      dtype: string
    - name: Medium
      dtype: string
    - name: Museum
      dtype: string
    - name: Museum-based instance ID
      dtype: string
    - name: Subset
      dtype: string
    - name: Width
      dtype: float64
    - name: Height
      dtype: float64
    - name: Product size
      dtype: float64
    - name: Aspect ratio
      dtype: float64
  config_name: default
  splits:
    - name: train
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/dataset.csv
    download_mode: reuse_dataset_if_exists
pretty_name: MAMe Dataset
size_categories:
  - 10K<n<100K
task_categories:
  - image-classification
tags:
  - image
  - artwork
  - museum

MAMe Dataset: Museum Artworks Medium

The MAMe Dataset is an image classification dataset focused on the recognition of mediums in artworks and heritage held by museums (e.g., Oil on canvas, Bronze or Woodcut).

The classes considered in the MAMe dataset comprise a wide variety of mediums according to both interpretations of the term. These can range from simple material aspects (e.g., Bronze, Silver or Gold) to complex, high-level techniques (e.g., Faience, Woodblock or Woven fabric). The variety of relevant features in MAMe requires both attention to detail and to the overall image structure.

---

Paper

• Journal Version: Materials in Art and Museum Environment (MAMe): A Dataset for Art Material Recognition
• ArXiv Version: MAMe: A Dataset for Multi-class Classification of Materials in Artworks

---

Dataset Variants

• MAMe_small: The images compressed in data/images/small.zip form a toy version of the dataset, optimized for quick experimentation and reduced storage needs.
• MAMe_original: The full version of the dataset, intended for tasks that require high precision in material classification. It consists of 205 GB of images. The entire dataset can be downloaded by running the following command or by clicking on this link:

  wget https://storage.hpai.bsc.es/mame-dataset/MAMe_data.zip

---

Dataset Description

The MAMe dataset contains thousands of artworks from three different museums, and proposes a classification task consisting on differentiating between 29 mediums (i.e. materials and techniques) supervised by art experts.

• Curated by: HPAI
• License: The MAMe dataset is available for non-commercial research purposes only.

Citation

If you use this dataset, please cite the following journal paper:

@article{pares2022mame,
  title={The MAMe dataset: on the relevance of high resolution and variable shape image properties},
  author={Par{\'e}s, Ferran and Arias-Duart, Anna and Garcia-Gasulla, Dario and others},
  journal={Applied Intelligence},
  volume={52},
  number={12},
  pages={11703--11724},
  year={2022},
  publisher={Springer},
  doi={10.1007/s10489-021-02951-w}
}

For accessibility purposes, you can also reference the ArXiv version:

@article{pares2020mame,
    title={The MAMe Dataset: On the relevance of High Resolution and Variable Shape image properties},
    author={Par{\'e}s, Ferran and Arias-Duart, Anna and Garcia-Gasulla, Dario and Campo-Franc{\'e}s, Gema and Viladrich, Nina and Labarta, Jes{\'u}s and Ayguad{\'e}, Eduard},
    journal={arXiv preprint arXiv:2007.13693},
    year={2020},
    url = {https://arxiv.org/pdf/2007.13693}
} 

---

Dataset Card Authors

Ferran Parés, Anna Arias-Duart, Dario Garcia-Gasulla

Dataset Card Contact

For more information or questions about this dataset, please contact the HPAI organization.