How does Agit generate synthetic data?
Agit uses 3D software to create massive human motion datasets, preserving biomechanical accuracy through an extensive library of professional motion capture (mocap) data. This allows for generating thousands of video sequences with varied body types, camera angles, lighting, environments, and clothing.
What data formats and labels are included in Agit's synthetic datasets?
Agit provides ready-to-train synthetic datasets with RGB video sequences, pixel-perfect depth maps, segmentation masks, sparse 2D/3D keypoints (COCO, MediaPipe), and dense keypoints. Detailed character information like skin color and anthropometric measurements are also included.
Is Agit's data GDPR/HIPAA compliant?
Yes, Agit's data is 100% synthetic and anonymous, containing no Personally Identifiable Information (PII). This eliminates all privacy and compliance risks associated with GDPR, HIPAA, and other regulations, allowing for secure model training and deployment.
Does Agit's library include common exercise postural errors?
Yes, Agit's library includes many common postural errors for a wide range of exercises. This allows AI models to not only recognize exercises but also accurately correct a user's form, addressing subtle nuances that impact user condition and performance.
How does Agit handle body occlusion in its data?
Unlike human annotators, Agit's pipeline has access to the accurate position of every body part. This enables the provision of precise keypoints even when a limb is fully occluded by the body or other objects, offering 'superhuman' ground truth for model training.