Software Developer

• Feb 2022 - Present

• Work on person/head detector improvement (testing new neural networks, augmentations, train dataset refinement). Pushed F1-score by 4.5% from 85% to 89.5%, eliminated most frequent false detections. • Work on products detection in retail stores (using SKU110K dataset for training and manually collected dataset for fine-tuning). Reached 95% F1-score on test dataset. • Work on products embedder model to recognize coffee products based on customer's products database (embedder working together with products detector). • Work on weapon detector improvement (testing new networks and augmentations, train dataset refinement, elimination of false positive predictions). Pushed F1-score by 6% from 84% to 90% on the test dataset. • Work on pricetag detection and recognition service. Recognition included OCR for product name, price and special offer price. • Development and support of real-time video analytics platform. • Work on local GitLab, infrastructure and services, including: - wrapping of various CV networks and models in services to work with the video analytics platform - making installation scripts for services in order to compile them into binaries, including virtual environment creation, dependencies installation, compilation of additional components - neural networks storage and distribution services, including model updates, license checking, etc. • Refactoring of existing services, converting Python services into Golang to boost inference speed and RPS. Show less

Junior Software Developer

• Apr 2021 - Feb 2022

- Development of videoanalytics pipelines, including RGB-cameras and thermocameras - Develompent of microservices (python/Flask, Go/Gin-Gonic), including: -- detection - Yolo, MobileNet -- distance estimation - custom model based on two MobileNets and object coordinates/size as input -- services for model storage, versioning, encryption and automatic upload/download (using own FTP-server and self-hosted Amazon S3) - Processing of different scenarios in videoanalytics (counting objects in a frame/in an area, object tracking, screnarios of crossing a line or entering a zone) - R&D of existing detection models/architectures, customizing and optimizing model in order to reduce size and increase inference speed; the final goal is to run detection on Coral Edge TPUs - Models converting (darknet, pyTorch, TF/tflite) - Working with hardware, including embedded devices, single-board computers, IP-cameras, thermocameras, lidars Show less

Deputy Head of Engineering Department

• Oct 2013 - Apr 2021

Engineer of Engineering Department

• Apr 2010 - Oct 2013