Sign in to view source links and access this dataset
Description
A synthetic video dataset containing 15 hours of human motion sequences, created by Intelligent-Systems. It includes 10,450 sequences at 30fps, totaling 1.6 million images, along with ground truth data and segmentation masks. The dataset was mirrored to Hugging Face and last updated on March 10, 2026.
Use Cases
Training human pose estimation models based on the provided camera and scene ground truth.
Developing video segmentation algorithms using the included segmentation masks.
Benchmarking depth estimation models based on the companion depth data mirror.
Generating synthetic training data for action recognition tasks using the 10,450 video sequences.
Strengths
Contains 1.6 million synthetic images across 10,450 sequences, providing substantial volume for training.
Includes multiple data modalities: video files, ground truth CSV/JSON files, and segmentation masks.
Offers high-resolution image sequences at 1280x720 pixels.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Data may reflect synthetic generation bias inherent to the source platform.
Provenance
Source
Intelligent-Systems, mirrored from the BEDLAM project site.
Collection Method
Synthetically generated video data.
Freshness
Last updated 2026-03-10 09:49:48; freshness should be verified.
Access to additional information and data requires registration at the original project site.