Fitness AI: How synthetic data powers better workouts

Can AI-driven fitness programs, developed with synthetic dataincrease your exercise?

During the COVID-19 pandemic, home fitness programs were all the rage. From January to November 2020, approx 2.5 billion in health and fitness programs downloaded all over the world. This trend has been maintained and shows no signs of slowing down new information It predicts growth from $10 million in 2022 to $23 million by 2026.

As more people use fitness apps to train and track their progress and performance, fitness apps are increasingly turning to AI to power their offerings, incorporating technologies including AI-based exercise analysis, computer vision, human pose estimation, and natural language processing. uses.

Founded in 2018, Tel Aviv-based Datagen claims to provide “high-performance synthetic data with a focus on data for human-centric computer vision applications.”

The company just announced a new Smart Fitness domain with self-serve visuals synthetic data A platform that helps AI developers produce the data they need to train smart fitness equipment to analyze and “see” people exercising.

“At Datagen, our focus is on helping computer vision teams and accelerating their development of human-centric computer vision tasks,” Datagen CEO Ofir Zuk told VentureBeat. “Almost every use case we see in the AI ​​space involves a human. We specifically seek to address and help understand the interactions between humans and their environment. We call it a person in context.”

Synthetic visual data represent fitness environments

The Smart Fitness platform provides 3D annotated synthetic visual data in the form of videos and images. This visual data can be used for body landmark estimation, pose analysis, posture analysis, repetition counting, object identification, etc. accurately represent fitness environments, advanced motion, and human-object interactions for tasks related to

In addition, teams can use the solution to generate whole-body motion data to iterate on their model and quickly improve its performance. For example, in pose estimation analysis situations, an advantage the Smart Fitness platform provides is the ability to quickly simulate different camera types to capture different differentiated exercise synthetic data.

Challenges in training AI for fitness

Pose estimation, a computer vision technique that helps determine the position and orientation of the human body with a human image, is one of the unique solutions offered by artificial intelligence. It can be used in avatar animation for artificial reality, such as markerless motion capture and worker pose analysis.

In order to properly analyze the posture, it is necessary to take several pictures of the human actor with the interaction environment. A trained convolutional neural network then processes these images to predict where the human actor’s joints are located in the image. AI-based fitness apps generally use the device’s camera, recording videos at up to 720p and 60fps to capture more footage during your workout.

The problem is that computer vision engineers need a lot of visual data to train AI for fitness analysis when using a technique like pose estimation. Data on how people perform various forms of exercise and interact with many objects is quite complex. Data should also be sufficiently varied to avoid high variability and bias. Collecting accurate data covering such diversity is nearly impossible. Moreover, manual annotation is slow, prone to human error, and expensive.

While a reasonable level of accuracy has already been achieved in 2D pose estimation, 3D pose estimation lacks in terms of generating accurate model data. This is especially true for drawing conclusions from a single image and without any in-depth information. Some methods use multiple cameras aimed at a person, receiving information from depth sensors to get better predictions.

However, part of the problem with 3D pose estimation is the lack of large annotated datasets of people in outdoor environments. For example, large datasets for 3D pose estimation Human 3.6M shot entirely indoors to eliminate visual noise.

There are ongoing efforts to create new data sets with more diverse information about environmental conditions, clothing diversity, strong articulations, and other influencing factors.

Synthetic data solution

To overcome such problems, the tech industry now makes extensive use of synthetic data, a type of artificially produced data that can closely mimic operational or production data, to train and test AI systems. Synthetic data offers several significant advantages: It minimizes restrictions on the use of regulated or sensitive data; can be used to customize data to fit conditions that real data does not allow; and allows for large training datasets without requiring manual data labeling.

according to report By Datagen, the use of synthetic data reduces production time, eliminates privacy concerns, provides reduced bias, annotation and labeling errors, and improves predictive modeling. Another advantage of synthetic data is the ability to easily simulate different camera types when generating data for use cases such as pose estimation.

Exercise demonstration simplified

With Datagen’s intelligent fitness platform, organizations can create tens of thousands of unique personalities performing different exercises in different environments and conditions – in a fraction of the time.

“With the power of synthetic data, teams can generate all the data they need with specific parameters in a matter of hours,” Zuk said. “This not only helps retrain the network and machine learning model, but also allows us to refine it in a short amount of time.”

In addition, he explained, the Smart Fitness platform optimizes your ability to access millions of meaningful visual training data, eliminating the repetitive burden of manually capturing each item.

“Through our constantly updated library of virtual human personalities and exercise patterns, we provide detailed pose data, such as the locations of joints and bones in the body, that can help analyze complex details to improve AI systems,” he said. “Adding these kinds of visual capabilities to fitness apps and devices can significantly improve the way we view fitness, allowing organizations to provide better services both in person and online.”

Fitness AI and synthetic data in the enterprise

According to eminent VP analyst Arun Chandrasekaran Gartnersynthetic data is, so far, “an emerging technology with a low rate of enterprise adoption.”

However, he says he will see increased adoption for use cases where data must be guaranteed to be anonymous or where privacy (such as medical data) must be preserved; increasing real data, especially where data collection costs are high; when there is a need to balance class distribution within existing training data (e.g. with population data) and emerging AI use cases where limited real data is available.

Several of these use cases are key to Datagen’s value proposition. When it comes to increasing the capabilities of smart fitness devices or apps, “increasing data quality, covering a wide gamut of scenarios, and preserving privacy during the ML training phase will be of particular interest,” he said.

Zuk acknowledges that it’s still early days for artificial intelligence and synthetic data, even digital technologies, to be brought into the fitness space.

“They are very non-reactive, very lean in terms of capabilities,” he said. “I would say that adding these visual capabilities to these fitness apps will definitely improve things significantly, especially as people exercise more in their own homes. We clearly see the demand growing and can imagine what people can do with our data.