Generate Synthetic Data for Analysis of People Exercising
Analyzing Human Performance in Fitness Environments
The Datagen Platform provides high quality, perfectly annotated data, in the form of video and images, that are used to train CV ML models for tasks related to understanding human behavior in fitness environments.
Synthetic Data for Human Analysis in Smart Fitness
In recent years, there has been a shift in how we stay fit, moving from gyms to fitness at home. More people are using fitness apps for training and monitoring their performance and progress. With AI on the rise, fitness apps are looking for smart solutions which can add value for their users in the form of AI based exercise analysis. To train AI for fitness analysis, computer vision engineers require vast amounts of visual data.
Data involving humans performing exercises in various forms and interacting with various objects is highly complex. It is also critical that the data is high-variance and sufficiently diverse to avoid bias. Collecting real data which covers this variety, as well as edge cases, is impossible. Additionally, manual annotation is slow, prone to human error, and expensive.
The Datagen Platform provides high quality, perfectly 3D annotated visual data in the form of video and images. This visual data has accurate representations of fitness environments, advanced motion and human-object interactions for tasks related to body keypoint estimation, pose analysis, posture analysis, repetition counting, object identification and more. Teams can use our solution to generate full-body in-motion data to quickly iterate on their model and improve its performance.
Use case examples
Identifying exercise types such as lunges, push-ups etc.
Counting the number of exercise repetitions
Identifying correct/incorrect form
Why Datagen for Smart Fitness
The right domain-specific data
Controllable camera devices
Controllable data generation
Frictionless granular control by the computer vision engineer
3D ground truth annotations and perfect 2D quality
Zero biases in the data distribution
Ability to define the distributions for every part of the data with no inherent biases