dtudk/MozzaVID_Large

MozzaVID dataset - Large split

A dataset of synchrotron X-ray tomography scans of mozzarella microstructure, aimed for volumetric model benchmarking and food structure analysis.

[Paper] [Project website]

This version is prepared in the WebDataset format, optimized for streaming. Check our GitHub for details on how to use it. To download raw data instead, visit: [LINK].

Dataset splits

This is a Large split of the dataset containing 37 824 volumes. We also provide a Base split (4 728 volumes) and a Small split (591 volumes).

!!! The above numbers are not exact at the moment as we have not released the test dataset yet.

Citation

If you use the dataset in your work, please consider citing our publication:

@misc{pieta2024b,
  title={MozzaVID: Mozzarella Volumetric Image Dataset},
  author={Pawel Tomasz Pieta and Peter Winkel Rasmussen and Anders Bjorholm Dahl and Jeppe Revall Frisvad and Siavash Arjomand Bigdeli and Carsten Gundlach and Anders Nymark Christensen},
  year={2024},
  howpublished={arXiv:2412.04880 [cs.CV]},
  eprint={2412.04880},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2412.04880},
}

Visual overview

We provide two classification targets/granularities:

• 25 cheese types
• 149 cheese samples

Fig 1. Overview of slices from each cheese type, forming the 25 coarse-grained classes. Fig 2. Example slices from the fine-grained classes. Each row represents a set of six samples from one cheese type (coarse-grained class), forming six consecutive fine-grained classes.