MM-CAD:A

MM-CAD:A MM-CAD:A is a multi-modal CAD dataset of 33,816 engineering models consolidated from 11 public 3D shape benchmarks, with every model aligned across five modalities and given natural-language captions. It is the encoder-fine-tuning half of the MM-CAD dataset released with our SGP 2026 paper. The much larger, automatically-curated MM-CAD:B (192,626 ABC models) is released separately; the two parts are disjoint — MM-CAD:A trains the encoders that make the MAAS curation of… See the full description on the dataset page:

Source: Hugging Face Hub (exanos/MMCAD). Metadata imported from the dataset’s Hub tags.