pedroodb/glosl-isharah

GloSL — Isharah (Saudi Sign Language) — continuous SLR/SLT

Part of GloSL, a set of continuous sign language translation corpora in one format.

• Sign language: Saudi Sign Language (sdl)
• Spoken language: Arabic (ar)
• Domain: multi-topic sentences, smartphone recordings in varied scenes
• Samples: 15,000 (37.52 h) — splits {'train': 13500, 'dev': 750, 'test': 750}
• Split policy: Isharah official Task-2 unseen-sentence split (Al Yami, Luqman et al., 2025)
• License: dataset terms (snalyami.github.io/Isharah_CSLR, added 2026-09-28): may not be redistributed, republished, sublicensed or hosted elsewhere without prior written permission; the paper is CC BY-NC-ND 4.0 — public; source: https://snalyami.github.io/Isharah_CSLR/
• Access: Poses published by GloSL on 27 Oct 2026, pending the authors' answer to a permission request of 7 Oct 2026. The repository is made private if the corpus's authors ask.
• Cite: Sarah Al Yami, Hamzah Luqman, et al., Isharah: A Large-Scale Multi-Scene Dataset for Continuous Sign Language Recognition, 2025 (arXiv:2506.03615)

Files

annotations.csv, card.json, poses/<NNNN>/<id>.pose. card.json holds the full provenance.

annotations.csv columns (same in every GloSL dataset; empty when not applicable): id, split, text, gloss, gloss_type, signer, start, end, fps, n_frames, video_ref, pose_file, video_file.

Poses

authors' keypoints mapped to mediapipe-holistic-543, 543 points (POSE_LANDMARKS 33, FACE_LANDMARKS 468, LEFT_HAND_LANDMARKS 21, RIGHT_HAND_LANDMARKS 21), 2D. Coordinates: pixels of the frame the pose refers to (header width x height, depth 0); z raw. Confidence: body = MediaPipe visibility (1.0 where the source shipped none); face/hands 1 when detected; missing point = (0,0,0) with confidence 0. Frame rate: native, per file (also in annotations.csv). Normalization (glosl-norm-v1) is applied at load time, not stored. Note: author-provided 86-keypoint poses (hands+lips+body) mapped into the corpus-standard 543 layout: body 25 -> POSE[0:25] (exact), hands -> standard hand components (exact), lips 19 -> MediaPipe outer-lip-contour FACE indices (per-lip face-mesh ordering not published by source; placement is within the lip region and does not affect topology, labels, or shoulder-based load-time normalization). Lower-body POSE[25:33] and non-lip FACE points are absent (confidence 0). Pixel-space coords. Header width/height are INFERRED per clip (smallest standard phone frame, portrait or landscape, holding the keypoints): the source video resolution is not published.