MUCCA (Multi-Category Class-Agnostic Counting) Dataset: A Collection of Multi-Category Images for Class-Agnostic Object Counting

The dataset

The MUCCA (Multi-Category Class-Agnostic Counting) dataset is a collection of 200 images specifically designed for testing open-world text-specified class-agnostic counting methods, with the distinctive feature of including multiple object categories per image. Class-agnostic counting (CAC) aims to count instances of arbitrary object classes beyond the categories encountered during training. Among the different possibilities to specify the target object category to count, the open-world text-guided paradigm expresses it as a textual description.

Images were collected from the web using Google search queries involving multiple object categories that reflect realistic contextual and semantic dependencies, e.g., ``images of markets with many berries''.  Subsequently, each image was manually annotated by placing a dot at the most representative point of every object instance belonging to the target classes, following annotation practices commonly adopted in class-specific and CAC benchmarks. Dot annotations were associated with specific object categories, resulting in category‑specific sets of dots. In addition to dot annotations, we also provide the corresponding class names, which can be directly used as textual queries for open‑world text‑guided CAC methods.

Images are stored in the images folder. Annotations consist of two .json files:- annotations.json --> for each image, it provides the (x, y) coordinates of the annotated dots along with the corresponding class ID.- class_mapping.json --> it defines the mapping between class IDs and class names.

Citing our work

If you found this dataset useful, please cite it

@dataset{mucca_dataset,
    author = {Ciampi, L., Pacini, G., Messina, N., Amato, G., & Falchi, F.},
    title = {MUCCA (Multi-Category Class-Agnostic Counting) Dataset: A Collection of Multi-Category Images for Class-Agnostic Object Counting},
  month = march,
  year = 2026,
  publisher = {Zenodo},
  version = {1.0.0},
  doi = {10.5281/zenodo.19231375},
  url = {https://doi.org/10.5281/zenodo.6560823}
}

Contact Information

If you would like further information about the dataset or if you experience any issues downloading files, please contact us at [e-mail], [e-mail], [e-mail]