Final mapping logic adjusted: image_id from highlighted captions matches person_idx + 1. order_in_scene recorded. (SUBSET: None) (part 00001-of-00002)

Added detailed pose annotations with Base64 encoded images, for lazy decoding (Full Dataset). (part 00001-of-00002)

Upload dataset (part 00001-of-00002)