cv

General Information

Full Name Andrei Popov
Title ML Engineer
Languages English, Russian

Interests

  • Computer vision, deep learning, natural language processing, ml ops

Experience

  • aug 2019 - present
    ML Engineer, Napoleon IT

    • Successfully enhanced the speed and quality of the retail store's product and price recognition system, resulting in a 20% improvement in accuracy. (PyTorch, TensorFlow, OpenVino, Flask, XGBoost, Hydra, Docker, Metric Learning)
    • Development and deployment of a computer vision module for detecting irregularities in production (PyTorch, Nvidia Triton Server, TensorRT, YoloV5, DeepSort, EfficientNet, Docker)
    • Development and deployment of a system for monitoring and controlling access to the office by face with an additional module of temperature analysis (PyTorch, RetinaFace, Docker)
    • Development of a system that counts unique visitors to the store through the use of a video camera (PyTorch, OpenCV, Docker)
  • 2022
    Mentor, AI Talent Hub

    • Providing expert guidance to students on deep learning and computer vision-related matters
    • Conducting thorough evaluations of students' final project submissions

Education

Projects

  • 2022
    Educational Video Analytics,

    • The system designed for use in educational institutions. It can recognise students' emotions in real-time through video. The following technologies were used in developing the system: face detector (RetinaFace), face tracker (DeepSort), emotion classifier (Deep Alignment Network).
  • 2022
    Landmark Retrieval,

    • A service for searching for similar landmarks based on a query image was developed. The training utilized a metric-learning approach. The final search is performed using FAISS.
  • 2022
    Music Track Artist Search,

    • A collection of anonymized acoustic features from music recordings has been provided. The objective is to identify the most suitable authors. In addition to classification, the model can also be used to evaluate the similarity between different artists and to distinguish different artists but with similar names.
  • 2022
    Aerial Photos Matching using Deep Learning,

    • A model was developed that takes an image as input and determines its position and rotation angle on the substrate. The challenge lies in the possibility of the substrate and image being captured at different times of the year or with overlapping clouds.
  • 2021
    Voice Commands Recognition,

    • Various models were compared for classifying different voice commands, and the ensemble of ResNet18 and EfficientNetB0 showed the best performance. Mel-spectrograms were used as input features.
  • 2020
    Speaker Recognition Bot,

    • A Telegram bot to remember and people by their voice. Speaker recognition is achieved by extracting unique voice features, storing them in a database, and using the K-Nearest Neighbors classifier to assess vector similarity during the inference stage.

Publications

  • 2021
    • Popov, Andrey S. and Ivanov, Sergey A. Neural Network Models for Russian Language Speaker Recognition. (2021) 2021 International Conference on Quality Management, Transport and Information Security, Information Technologies (IT&QM&IS)
  • 2022
    How computer vision helps determine the coordinates of satellite images, article on Habr.com

    • This article covers how to won the second place in the Digital Breakthrough contest with a solution for automating photo georeferencing. The main insight is that basic solutions are not always good and proven.

Competitions and Awards