CFM is a language-aligned concept foundation model that extracts spatially localized, visually grounded, and human-interpretable concepts at various granularities from images, organizes them into hierarchies, and automatically assigns names to them — enabling concept-based explanations for any downstream task that the foundation model can perform, such as classification, open-vocabulary segmentation, and captioning.

@article{wittenmayer2026cfm,
  title   = {CFM: Language-aligned Concept Foundation Model for Vision},
  author  = {Wittenmayer, Kai and Rao, Sukrut and Parchami-Araghi, Amin
             and Schiele, Bernt and Fischer, Jonas},
  journal = {arXiv preprint arXiv:2601.13798},
  year    = {2026}
}

Concepts and their names are extracted automatically from large-scale image data without manual curation. Some concepts or their automatically assigned names may be inaccurate or offensive; they do not reflect the views of the authors.