Standardized behavioral segmentation datasets

A growing, standardized collection of behavioral-segmentation datasets: recordings in which behavior has been parsed into a timeline of labeled segments (an ethogram), stored in a single common format so that behavior-analysis tools can be developed and benchmarked across datasets without a separate loader per dataset. Each file uses the ndx-ethogram NWB extension (an EthogramBouts table plus an Ethogram catalogue) with markerless pose estimation (ndx-pose) and source video where available. The collection spans unsupervised motif/syllable discovery (keypoint-MoSeq, B-SOiD, VAME, and others via the Mlost benchmark) and dense expert human annotation (CalMS21). These are curated, derived conversions of publicly available datasets redistributed under CC-BY 4.0 with attribution to the original authors (keypoint-MoSeq: doi.org/10.5281/zenodo.10636983; Mlost: doi.org/10.5281/zenodo.14382973; CalMS21: doi.org/10.22002/D1.1991).