Synthetic Data with 2D & 3D key-point
Synthetic Data with
2D & 3D key-point
Dynamic data refers to synthetic data for objects with actions like humans or animals.
Accurate action data can be collected with a high-precision motion capture system,
and dynamic synthetic data can be created by integrating a static data generation system.
Workflow
Motion Capture
Utilize high-precision motion capture systems for capturing raw data.
Refining
Finalize raw data adjustments and remove noisy data.
3D Modeling
A variety of character backgrounds can be modeled in 3D
Skinning
3D models combined with motion data.
Renderring
Creating high-quality source data.
Auto Labeling
Various labeling data can be generated automatically.
3D Motion Capture (Source) Data
It is important to collect high-quality source data for accurate key point labeling of dynamic data. In this stage, a high-precision optical motion capture system and a high-resolution camera are used. In the case of human pose, the coordinates of 2D & 3D key points in the view required for AI training will be extracted and used as source data.
3D Modeling
Diversity is key to the development of an AI model. The 3D model resembles real data (clothes, background). 3D modeling completely overcomes the limitations of the inaccuracy of existing 2D video labeling and can be expected to have an impact in terms of personal information protection.
Synthetic image key point labeling
Skinning and rendering of 3D skeleton data collected through a 99.9% high-precision motion capture system on top of the 3D model to obtain customized key point labeling data with various views and domains.