Engineering teams built around your needs — from solo engineers to full-scale squads.
SOFTWARE DEVELOPMENT Custom Software Development Web Application Development Backend Development Frontend Development E-commerce Development MVP Development SaaS Development UX/UI Design AI & MOBILE Machine Learning Deep Learning Computer Vision Natural Language Processing Agentic AI LLM Integration iOS App Development Android App Development Cross-Platform App Development CLOUD, QA & DATA Cloud Migration Cloud Architecture DevOps & CI/CD Cybersecurity Software Testing & QA Automation Testing Business Intelligence Big Data CRM All Services → OUR PRODUCTSSell more in every chat Support that never sleeps Every delivery in the loop Campaigns people answer
Product offers, cart reminders and checkout — right inside Viber, where your customers already are.
Boost my salesPrecise Microclimate Remote Control Full Traceability Autonomous Operation
Automated management of temperature, humidity, lighting, and ventilation for reliable breeding of beneficial bioagents.
Request a DemoWe've delivered 60+ projects across 11+ sectors. Experience in your industry? We have it.
Agriculture Banking Fintech Insurance Healthcare Real Estate Logistics IoT AI Cybersecurity Blockchain All Industries → Careers Technologies Our Work BlogMMS.
A mobile/embedded machine learning model for real-time background subtraction of video streams, plus the end-to-end infrastructure to train, version and distribute models to devices.
AI · Mobile computer visionReal-time video ML, running on the device.
The goal of this project was to develop a mobile/embedded Machine learning model and training infrastructure for model versioning. The model binary was designed to capture video streams from the mobile device, perform background subtraction and have the person remaining in the video, and apply shader functions for the background areas.
The whole infrastructure was managed by an end-to-end platform for deploying to production and development of ML pipelines and models. Also distribution and updates of models to the embedded devices was ensured.
On-device pipeline Capture the video stream Background subtraction, keeping the person Shader functions for the background Mobile GPU / DSP Quantized model, real timeSmall enough for the phone, fast enough for live video.
The challenge of the project was the converting and post training quantization of the model to ensure correct performance of the model on the embedded/mobile device, and also ensure execution on the GPU and/or DSP of the mobile device.
01Model conversion for embedded/mobile devices
02Post training quantization with correct performance
03Execution on the GPU and/or DSP of the mobile device
From research to devices in the field.
The client assigned Generic Soft with the training and quantization of machine learning computer vision algorithms for real time background subtraction of video streams, including model research and design.
Model research and design
Full ML pipelines, from data capturing to GPU/DSP preparation
Deployment of infrastructure servers for version control of models and pipelines
