CLERC Épée v0.2: Sign Language Data Layer
收藏资源简介：
CLERC Épée v0.2 is a multi-signer pilot dataset of American Sign Language (ASL): 600 clips of 150 parallel phrases signed by 4 Deaf signers (150 clips each), in a fully parallel structure where every phrase is signed by all four signers. Each clip is represented as a (n_frames, 128, 3) keypoint array combining MediaPipe Hands, Pose, Face (eyes + mouth), and FACE_OVAL head silhouette, with privacy-preserving design (no internal facial features beyond eyes and mouth). Annotations include gold-standard ASL gloss segmentation with temporal boundaries. New in v0.2: a fourth signer, corpus expanded from 300 to 600 clips, fully parallel phrase structure, and a cross-signer generalization benchmark (BENCHMARK.md): accuracy on a held-out signer rises from 22% to 59% (macro-F1 0.13 to 0.38) as training signers increase from 1 to 4. The dataset is mirrored on Hugging Face at CLERC-DATA/epee and released under CC BY-NC-SA 4.0. Commercial licensing and full corpus access: contact florian@clerc.io. Statistics: 600 clips, 69,504 frames, 38.61 min signed duration, 1,708 gloss tokens, 251 unique glosses, 99.98% MediaPipe head-silhouette detection rate.
