This repo packages the IARPA Janus Benchmark-A (IJB-A) face recognition dataset
in its CleanData layout, plus the official 10-split 1:1 verification and 1:N
identification protocols.
Unlike LFW-style benchmarks, IJB-A is template-based: each subject is
represented by a template aggregating multiple still images and/or video
frames. Protocol CSVs map every (template, file) row to a face annotation with
bounding box, landmarks, and demographic attributes.… See the full description on the dataset page:
https://huggingface.co/datasets/marcelohaps/ijb-a.