kubricount-subset is a FiftyOne image dataset built from a subset of
liuchang666/KubriCount,
a large-scale synthetic benchmark for multi-grained visual counting. In
addition to the ground-truth counting supervision, this FiftyOne version ships
open-vocabulary detection predictions from NVIDIA's LocateAnything-3B, SigLIP
embeddings, and per-sample counting evaluation fields.
If you haven't already, install… See the full description on the dataset page:
https://huggingface.co/datasets/harpreetsahota/kubricount-subset.
