Sample Vendors: Anthem; Gretel Labs; Statice, V7
Range: Long (6 to 8 Years)
Image and video synthetic data is three to six years away from early majority adoption because of current lack of customer understanding for the use and application of synthetic data. Vendors continue to educate and provide guidance throughout the model training process for the successful adoption of image synthetic data. The supply side is also nascent. Venture capital (VC) funding for image and video synthetic data generation technology startups totals only $69 million in the last five years, with an uptick in investment growth to $24 million in 2020 (from $4 million in 2019). However, in 2021, investment remains flat at $25 million.
Common synthetic images and videos offerings available today are focused toward other industries and are not specific to insurance, such as:
Object detection and object classification in manufacturing and retail ■
Geospatial analytics from satellite imagery, aerial imagery, drone imagery, among others ■
Simulation environment with sensor information for training autonomous navigation systems and advanced driver assistance systems (ADASs) ■
Human facial data for driver state in-cabin monitoring, facial analytics such as emotion analysis, attention analysis and so on ■
Indoor/outdoor digital content for immersive technology (AR/VR) ■
However, each of these have applicability either to train insurer’s own risk models, or to provide greater accuracy to external models that an insurer connects to.
The market is in an early stage of adoption. It is difficult for the vendors to provide a self- service platform to generate synthetic data for model training, where data is insufficient or privacy is restricted. Although self-serving customers is the vision of most vendors, image and video synthetic data is most usually provided as a service.
Mass: High
Gartner, Inc. | G00786204
Page 37 of 48
This research note is restricted to the personal use of abhishek.sharma@fractal.ai.
Made with FlippingBook - PDF hosting