pedroodb/glosl-phoenix14t

GloSL — RWTH-PHOENIX-Weather 2014T

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

• Sign language: German Sign Language (DGS) (gsg)
• Spoken language: German (de)
• Domain: weather forecast interpreting, German public TV
• Samples: 8,257 (10.53 h) — splits {'train': 7096, 'dev': 519, 'test': 642}
• Split policy: glosl-splits-v1, imported — RWTH-PHOENIX-Weather 2014T official train/dev/test (Camgöz et al., CVPR 2018) (see card.json and diagnostics/)
• License: CC BY-NC-SA (verify 3.0 vs 4.0 on the badge) — public; source: https://www-i6.informatik.rwth-aachen.de/~koller/RWTH-PHOENIX-2014-T/
• Cite: Camgöz et al., Neural SLT, CVPR 2018

Files

annotations.csv, card.json, poses/<NNNN>/<id>.pose, videos/<NNNN>/<id>.mp4 (same frames as the 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

mediapipe-holistic, 543 points (POSE_LANDMARKS 33, FACE_LANDMARKS 468, LEFT_HAND_LANDMARKS 21, RIGHT_HAND_LANDMARKS 21), 3D. 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: reused from slt_datasets MediaPipe-Holistic .npy (KB/12); normalization applied at LOAD time (KB/05)