MEDIAPI-SKEL is a 2D-skeleton video corpus of French Sign Language (LSF) with French subtitles. To constitute this corpus, we used 368 subtitled videos totaling 27 hours of LSF footage and written French produced by Média'Pi (https://media-pi.fr/), a media company producing bilingual content with LSF and written French . From the original videos, we extracted 25 body keypoints, 2x21 hand keypoints and 70 face keypoints using OpenPose (https://github.com/CMU-Perceptual-Computing-Lab/openpose). We provide the 135 keypoints for every person in every frame of the 368 videos. We also provide the associated subtitles in French. We cannot provide the original videos, but they can be accessed through a subscription with Média'Pi. The subtitles of this corpus are accurately aligned to the 2D-skeleton video content. Each subtitle corresponds to the associated segment of Sign Language video. Data 2D-skeleton data for each frame of each of the 368 videos in Json format Subtitles in written French in WebVTT format Resolution, frame rate and duration of each video Train/dev/test set separation

Creator(s)

LIMSI-CNRS - Annelies Braffort

LIMSI-CNRS - Michèle Gouiffès

hannah bull

Distributor(s)

LIMSI-CNRS - Annelies Braffort

LIMSI-CNRS - Michèle Gouiffès

hannah bull

Right Holder(s)

LIMSI-CNRS - Annelies Braffort

LIMSI-CNRS - Michèle Gouiffès

hannah bull