📚 Institutional Books: Harvard Library — Visual Elements
22 million visual elements extracted from the volumes that comprise the Institutional Books: Harvard Library dataset.
22,622,060 visual elements extracted from 983,004 volumes
766,992,447 o200k_base tokens in AI-generated captions
6 high-level classes of visual elements organized in splits
5 processing steps: Detection, Classification, Deduplication, Captioning, and Rotation
