- Owing to the pandemic, companies are forced to take up synthetic data as an option.
- The company plans to launch more APIs to address other computer vision challenges.
Introduction:
Self-driving vehicle companies spend millions of dollars every year to collect and label data according to an estimate. The third-party contractors list out hundreds of thousands of human data labeling. It will trace the annotations that machine learning models need to learn. The action of these data sets to include the correct distribution and frequency of samples has become exponentially difficult as performance requirements have grown. The pandemic has underscored how vulnerable these practices are as contractors have been forced to work from home. This has prompted companies to use synthetic data as an option.
Synthesis AI’s platform uses generative machine learning models, image rendering, and composition, and various other methods to develop and label images of objects, scenes, people, and the environment. The customers can change things like geometries, texture, lighting, image modality, and camera location to produce various results data to train computer vision models.
How Synthesis AI works?
Synthesis AI provides data sets that contain 10000 to 200000 scenes of common use cases. These include head poses and facial expressions, eye gazes, and near-infrared images. But the company uniquely provides an API that develops millions of images of realistic faces captured from various angles in various environments.
Using the API the customers can submit a job in the clouds to synthesize as much as terabytes of data. According to the company, API covers thousands of identities that span gender, age group, ethnicity, and skin tone. Generate modifications procedurally to the phases to reflect a change in expression and emotion. This also includes head turns and features like head and facial hair.
The built-in style adds on subjects with accessories like glasses, sunglasses, hats, headphones, face masks, and other headwear. Various other controls to avoid adjustment in camera optic, lighting, and post-processing. Synthesis AI claims that the data is unbiased and perfectly labeled. But the jury is out on the representatives of synthetic data.
What does the research say?
In a study conducted last January, the researchers at Arizona State University concluded an observation. The system trained on the images of engineering professors. 93% were males and 99% white. The system amplified the data set’s existing bias where 80% of the professors were males and 76% were white.
On the other hand startups like Hazy and Mostly AI have developed methods to control the bias of data that can reduce harm. There is a study published by a group of Ph.D. candidates at Stanford University that also claims the same thing. The co-authors claim that their technique allows them to weigh certain features. These features are more important to develop a diverse set of images for computer vision training.
Despite the competition from startups like Datagen and Parallel Domain, Synthesis AI says write measure technology and Hans manufacturers use its API to develop model training and test data sets. Affectiva builds AI and claims that it can understand emotions by facial expressions and speech analysis.
Statement from founder and CEO of Synthesis AI:
“One of our teleconferencing customers leveraged synthetic data to create more robust facial segmentation models. By creating a very diverse set of data with more than 1,000 individuals with a wide variety of facial features, hairstyles, accessories, cameras, lighting, and environments, they were able to significantly improve the performance of their models,” founder and CEO Yashar Behzadi said.
“[Another one] of our customers are building a car driver and occupant sensing systems. They leveraged synthetic data of thousands of individuals in the car cabin across various situations and environments to determine the optimal camera placement and overall configuration to ensure the best performance.”
In the future 11-employee company, Synthesis AI plans to launch an additional API to address various computer vision challenges. “To reach widespread adoption, we need to continue to build out 3D models to represent more of the real world and create scalable cloud-based systems to make the simulation platform available on-demand across a broad set of use cases.”
Existing investors Bee Partners, PJC, iRobot Ventures, Swift Ventures, Boom Capital, Kubera VC, and Leta Capital contributed to San Francisco, California-based Synthesis AI’s seed round announced today.