mmscale/

mmscale-data

21,977 (image, scenario) pairs across 8,444 DALL·E-generated images, each labeled with a single mean_rating ∈ [1, 5] for moral acceptability and a single modality_label ∈ {text, image, both} indicating which modality the judgment hinges on. This is a simplified eval view — one row per (image, scenario) pair with one rating and one modality label. The underlying paired-annotation source (per- annotator ratings + modality votes… See the full description on the dataset page: https://huggingface.co/datasets/mmscale/mmscale-data.