Couldn’t attend Transform 2022? Check out all the top sessions in our on-demand library now! Look here.
Can AI-powered fitness apps built with synthetic data ramp up your training?
During the COVID-19 pandemic, home fitness apps were all the rage. From January to November 2020 approximately 2.5 billion health and fitness apps were downloaded worldwide. That trend has continued and shows no signs of slowing down, with new dates forecast growth from $10 million in 2022 to $23 million in 2026.
As more people use fitness apps to train and track their development and performance, fitness apps increasingly use AI to power their offerings by providing AI-based training analytics, using technologies such as computer vision, human posture estimation and natural language processing techniques.
Founded in 2018, Tel-Aviv-based Datagen claims to “deliver high-performance synthetic data, with a focus on data for human-centric computer vision applications.”
The company just announced a new domain, Smart Fitness, on its self-service, visual synthetic data platform that helps AI developers produce the data they need to analyze people and train smart fitness equipment to “see.”
“At Datagen, our focus is on helping computer vision teams and accelerate their development of human-centric computer vision tasks,” Ofir Zuk, CEO of Datagen, told VentureBeat. “Almost every use case we see in the AI space is human-related. We specifically try to resolve and help understand the interconnection between humans and their interaction with the surrounding environments. We call it human in context.”
Synthetic visual data represents fitness environments
The Smart Fitness platform provides 3D annotated synthetic visual data in the form of video and images. This visual data accurately depicts fitness environments, advanced movement, and human-object interactions for tasks related to body key point estimation, pose analysis, posture analysis, repetition counting, object identification, and more.
In addition, teams can use the solution to generate full-body in-motion data to iterate their model and quickly improve performance. For example, in the case of pose estimation analysis, the Smart Fitness platform provides the ability to quickly simulate different camera types to capture a variety of differentiated synthetic training data.
Challenges for training AI for fitness
Pose estimation, a computer vision technique that helps determine the position and orientation of the human body with an image of a person, is one of the unique solutions AI has to offer. For example, it can be used in artificial reality avatar animation, as well as markerless motion tracking and pose analysis of workers.
To correctly analyze the pose, it is necessary to capture several images of the human actor with its interactive environment. A trained convolutional neural network then processes these images to predict where the human actor’s joints are in the image. AI-based fitness apps generally use the device’s camera and record videos at up to 720p and 60fps to capture more frames during exercise performance.
The problem is that computer vision engineers need massive amounts of visual data to train AI for fitness analysis when using a technique like pose estimation. Data where people perform exercises in various forms and interact with multiple objects is very complex. The data should also have a high variance and be sufficiently diverse to avoid bias. It is almost impossible to collect accurate data covering such a variety. In addition, manual annotation is slow, prone to human error, and expensive.
While an acceptable level of accuracy in 2D pose estimation has already been achieved, 3D pose estimation is lacking in terms of generating accurate model data. This is especially true for inferences from a single image and without depth information. Some methods use multiple cameras aimed at the person and capturing information from depth sensors to make better predictions.
However, part of the problem with 3D pose estimation is the lack of large annotated datasets of people in open environments. For example, large datasets for 3D pose estimation such as: Human3.6M were captured entirely indoors to eliminate visual noise.
There is an ongoing effort to create new data sets with more diverse data on environmental conditions, clothing variety, strong articulations, and other influential factors.
The synthetic data solution
To solve such problems, the engineering industry is now widely using synthetic data, a type of artificially produced data that can closely mimic operational or production data, for training and testing artificial intelligence systems. Synthetic data offers several important benefits: it minimizes the limitations associated with using regulated or sensitive data; can be used to adapt data to circumstances that do not allow real data; and it enables large training data sets without manual data labeling.
According to an report by Datagen, using synthetic data reduces production time, eliminates privacy concerns, reduces bias, annotation and labeling errors, and improves predictive modeling. Another benefit of synthetic data is the ability to easily simulate different camera types while generating data for use cases such as estimating poses.
Exercise Demonstration Made Easy
With Datagen’s smart fitness platform, organizations can create tens of thousands of unique identities that perform a variety of exercises in a variety of 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,” says Zuk. “This not only helps to retrain the network and machine learning model, but also allows you to fine-tune it in no time.”
In addition, he explained, the Smart Fitness platform optimizes your ability to capture millions of substantial visual workout data, eliminating the repetitive burden of personally capturing each element.
“Through our constantly updated library of virtual human identities and types of exercises, we provide detailed pose information, such as locations of the joints and bones in the body, that can help analyze intricate details to improve AI systems,” he said. he. “Adding such visual capabilities to fitness apps and devices could significantly improve the way we view fitness, enabling organizations to provide better services, both in person and online.”
Fitness AI and synthetic data in the enterprise
According to Arun Chandrasekaran, VP Analyst at GartnerSynthetic data has so far been an “emerging technology with a low degree of business adoption”.
However, he says there will be increasing use of use cases that require data to be guaranteed to be anonymous or privacy protected (such as medical data); augmentation of real data, especially when data collection costs are high; where there is a need to balance class distribution within existing training data (such as with population data), and emerging AI use cases for which limited real data is available.
Several of these use cases are critical to Datagen’s value proposition. When it comes to enhancing the capabilities of smart fitness devices or apps, “of particular importance is the ability to improve data quality, cover the wide range of scenarios and preserve privacy during the ML training phase,” he said.
Zuk admits it’s too early to bring AI and synthetic data, and even digital technologies in general, to the gym.
“They’re very non-reactive, very lean in terms of their capabilities,” he said. “I would say that adding these visual capabilities to these fitness apps, especially as people exercise more in their own homes, will definitely improve things significantly. We are clearly seeing an increase in demand and we can imagine what people with our data can do.”
The mission of VentureBeat is a digital city square for tech decision makers to gain knowledge about transformative business technology and transactions. Learn more about membership.