Synthetic counterfactual counting corpus for identifying an additive
neural-mass cardinality coordinate in vision-language models (NMCA).
Companion dataset to "Counting Requires Mass: An Algebraic and Causal Account
of Numerosity in Vision-Language Models."
Corpus version: masscount-cf-1.0.0
Master scenes: 99,989
Delivered images (this upload): 326,623
Object instances: 28,617,018
Scene graphs are the durable artifact; pixels are regenerable… See the full description on the dataset page:
https://huggingface.co/datasets/Ritabrata04/masscount-cf.