BIM-SIGN Videos: Malaysian Sign Language Raw Video Recordings

BIM-SIGN Video Data : Malaysian Sign Language Isolated-Sign Video Recordings BIM-SIGN (Video) provides the raw RGB video recordings underlying the largest isolated word-level Malaysian Sign Language (Bahasa Isyarat Malaysia, BIM) dataset assembled to date. It is the video companion of the BIM-SIGN dataset; a separate, non-identifying skeletal pose package (BIM-SIGN Pose, 15,258 samples) is deposited under open access on Zenodo for users who do not need, or should not have, access to face-visible video. The dataset covers 117 gloss classes spanning everyday communication (greetings, family terms, food, transport, question words, common verbs), comprising 15,277 video clips contributed by 220 signers, with 92 to 404 recordings per gloss (mean ~130). Recordings were captured with a Logitech C920 (1080p HD) webcam through a browser-based capture interface with a live MediaPipe pose/hand overlay, allowing signers to self-monitor their framing during recording. All participating signers provided informed consent for the release and controlled redistribution of their face-visible recordings. Each clip is provided as a single RGB video (MP4, with a small number of MOV files) showing one signer performing one isolated sign. Files are canonically named by gloss and sample index, with each sample's identifier shared with its corresponding pose sequences in the open BIM-SIGN Pose record, so the two can be merged by researchers with access to both. A seeded, stratified 80/10/10 train/validation/test split (seed 42) is provided, along with a full gloss vocabulary with English translations and per-gloss statistics. Access to this record is restricted. Because it contains identifiable imagery (signers' faces are visible), it is distributed under Restricted Access on Zenodo: access is granted on request, subject to a Terms of Use agreement prohibiting re-identification of signers, biometric or surveillance use, and commercial exploitation without prior permission from the authors. Documentation, sample visuals, and links to both the pose and video records are mirrored at the project page, https://bim-sign.github.io.