Datasetpublic

PragyaVLA/motionlang-1000-motions

MotionLang — 1000 curated G1 motions The 1000 unique motion clips used by the MotionLang language-centric browser (https://pragyaai.github.io/PragyaVLA). Balanced set: 10 categories x 20 clips x 5 language pages (English, Hindi, Bengali, Tamil, Telugu) = 1000 unique motions. Each *.npz is a G1-retargeted keypoint motion (base_frame_pos, base_frame_wxyz, joint_angles), sourced from PragyaVLA/PragyaVLA-G1-motions-dataset (HumanML3D/AMASS). File stems match the rendered video… See the full description on the dataset page: https://huggingface.co/datasets/PragyaVLA/motionlang-1000-motions.

sourceHugging Facemitupdated 1mo agoView on Hugging Face

0likes14downloads

Dataset Card

MotionLang — 1000 curated G1 motions

The 1000 unique motion clips used by the MotionLang language-centric browser (https://pragyaai.github.io/PragyaVLA). Balanced set: 10 categories x 20 clips x 5 language pages (English, Hindi, Bengali, Tamil, Telugu) = 1000 unique motions.

—Each *.npz is a G1-retargeted keypoint motion (base_frame_pos, base_frame_wxyz, joint_angles), sourced from PragyaVLA/PragyaVLA-G1-motions-dataset (HumanML3D/AMASS).
—File stems match the rendered video basenames in PragyaVLA/motionlang-motion-videos.
—manifest.csv maps every clip to its language, category, caption, English caption, clip_id, source npz path, and video basename.