epfl-vilab-modus/

MODUS-15Modality

MODUS is a large-scale, pixel-aligned 15-modality dataset for any-to-any multimodal training. Every sample aligns 15 modalities covering appearance, geometry, structure, segmentation, detection, text, and learned features. Paper: https://huggingface.co/papers/2607.25948 Code: https://github.com/EPFL-VILAB/Modus

	Modalities




	

Group Modalities

Appearance rgb, caption

Geometry depth, normal

Structure canny, sam_edge… See the full description on the dataset page: https://huggingface.co/datasets/epfl-vilab-modus/MODUS-15Modality.