AI Computer Vision
3D body (skeletal) tracking middleware
Facial and body recognition library for server, mobile and embedded solutions
Scalable API for facial
and body recognition
Advanced face liveness detection for digital onboarding and KYC
AI video analytics platform
for human activity recognition
Edge AI hardware
for face and body tracking
3DiVi Inc., founded in 2011, is one of the leading developers of AI and machine learning (ML) technologies for computer vision.

Core AI

Platform to create state-of-the-art machine learning models for computer vision
3DiVi provides customers with AI capabilities for computer vision in the form of "AI as a service" through our software platform. Core AI connects to customers' IT infrastructure via a set of standard interfaces and allows customers to use their own data to continuously train and update models.
Using Core AI to develop AI models is much more cost-effective than building your own AI-based computer vision infrastructure, which can take years and huge investments.

Key features

Quick start
Get your own ML model using automation tools - synthetic data generator, smart markup and AutoML.
Human in the Loop
Convenient tools that allow you to retrain the deployed ML model in the background using new data.
Federated Learning
Possibility of local or cloud deployment for cases that require confidentiality of training data.
Tech Support
Shorten your market entry time with the help of our Tech Support team!

Core AI includes 3 levels:

Base models provide a basis for the cost-effective production of commercial-scale ML models adapted to video and photo data processing.
3DiVi deep learning platform allows to create highly efficient ML models for computer vision, covering all stages from training to deployment and containing the following tools:
Core AI Platform is used together with 1.2 petaflops GPU cluster and custom storage. The system is optimized to store a huge amount of training data and supports millions of concurrent queries per second. 3DiVi continuously develops this infrastructure to provide computing resources and large-scale data management for training of powerful AI models at low cost.
Data Generation Tool
This tool generates synthetic training data which supplement real world data. Synthetic data in training has a number of benefits: no need for manual data markup, obtaining a large amount of data with relatively low labor costs, control over diversity of presented situations and magnitude of variations by choosing optimal generation parameters, simple expansion of data sets (via adding new targets and their parameters), data generation for cases that rarely occur in reality (for example, industrial accidents, fires, etc.).
Markup Tool
This tool marks up data for training and test sets in a semi-automatic mode. Thanks to intelligent interpolation technologies and other intelligent enhancements, the markup time is reduced by 5-10 times.
This tool is used to effectively train models, analyze errors, build metrics, graphs and visualize training results, including in dynamic mode on various subsets of data sets. Web-debugger allows to find the main types of errors, as well as subtle points in model operation. That helps to narrow down the range of hypotheses for testing and speed up the desired result.
Model Compression and Expansion Tool
Execution efficiency, memory usage and power consumption are critical factors in model deployment. To ensure portability to devices with harsh computing environments, we embedded model compression techniques such as quantization, reduction, and distillation and included them into the model production pipeline. Such methods are able to convert a trained model to a lightweight model that runs faster on edge devices and consumes less memory while maintaining comparable accuracy.
Continuous Learning Tool
The continuous learning tool allows to automate model retraining. The tool covers the whole life of the model from data entry to performance measurement in operation. Continuous learning ensures that your model is updated as soon as there are signs of degradation or changes in the environment.
Confidential Learning Tool (under development)
We integrate advanced confidential computing technologies to provide models to be trained with customer-side stored data. Transfer of raw data samples is excluded due to combination of computational encryption technologies. This ensures data security, privacy and regulatory compliance throughout the entire production process.
Learn more about Core AI Platform