Deep Learning

Recognition of Microcolonies in Bacterial Culture

Date: Jul 2020
Team: Solo
Role: CV Engineer
Language: Python 3.7
Architecture: UNet (with EfficientNet-B4 encoder)
Dataset: cups.mail
Stack: PyTorch, Catalyst, OpenCV, Pillow, Segmentation Models PyTorch, EfficientNet PyTorch, Pandas
GitHub: github.com/pandov/cups-mail

Description: The model was developed while participating in an online competition from
Digital Breakthrough. The model solves the problem of classification and segmentation.
The network builds a binary segmentation mask and classifies them into one of 6 different groups.
My best metric: 6.816 (max: 7.0)

Localization of the iris

Date: Oct 2020
Team: Art-Int Labs
Role: CV Engineer
Language: Python 3.7
Architecture: MobileNetV2 (modified)
Dataset: UnityEyes
Stack: PyTorch, TensorBoard, Catalyst, OpenCV, FilterResponseNormalizationLayer

Description: The model takes a single-channel eye image and returns
4 values: (X, Y) center point and (Width, Height) of the circumscribed rectangle.

3D Facial Landmarks

Date: Jun 2020
Team: Art-Int Labs
Role: CV Engineer
Language: Python 3.7
Architecture: Custom FCNN
Dataset: AFLW2000
Stack: Dlib, PyTorch, TensorBoard, Catalyst, OpenCV
Telegram: @pandovbot

Description: This solution uses a Neural Network model that approximates 2D points in 3D.
The Dlib library is used to recognize 2D facial landmarks. The model was trained on the AFLW2000 dataset.

Facial Expression Recognition - Telegram Bot

Date: Jun 2020
Team: Solo
Role: CV Engineer
Language: Python 3.7
Architecture: Custom CNN+FCNN
Dataset: FER2013, JAFFE, Cohn-Kanade, © RAF-DB
Stack: PyTorch, TensorBoard, Catalyst, OpenCV, Dlib, Pandas, DVC
GitHub: github.com/pandov/myitacademy
Telegram: @pandovbot

Description: This is my graduation project at Samsung IT Academy.

Android Development

YT Cinema

Date: Oct 2018
Team: Solo
Role: Android Developer
Language: Kotlin (Java)
Stack: Firebase, Data Binding, Room, Retrofit, Picasso, FragNav, Kotlin Coroutines, YouTube Android Player API

Description: Application for watching YouTube at the same time.
The video stream is synchronized for all devices connected to the session.

Game Modding

MTA DayZ Standalone

Date: Jul 2016
Team: VIME Team
Role: CTO
Language: Lua, SQL
Stack: mtasa-blue, MTA:SA API, MySQL
Game: MTA:SA

Description: The game mode is a copy of the DayZ Standalone game recreated on the
GTA San Andreas engine. During its existence, it has become very popular worldwide on the MTA platform.