Shakhzodbek
Yuldoshov
Computer Vision Engineer
Tashkent 100142
shahzodbek.yuldoshov@gmail.com
+998 90 139 13 31
Deep learning models used in my experience:
yolov5, yolov6, yolox, yolov7, yolov8, detectron2, unsupervised learning for image classification, knn
for digit detection and classification,
paddle ocr, yolov5 train on multigpu
Edge Devices:
• Jetson nano
• Jetson AGX
• esp32
camera
• OAK (OpenCV AI Kit cameras)
• Raspberry PI
Maximum work with big data:
3 TB of annotated images combined with 1.2 TB of annotated images. Trained state-of-the-art models
for traffic sign detection and segmentation tasks.
Data Annotation Types
• PascalVOC (bbox)
• COCO(polygon and bbox)
• labelme json format (polygon and bbox)
• custom json formats (polyline, polygon, bbox)
Tracking Algorithms:
• byte track with yolov5
• sort track with yolov5
•
custom tracker
Sponsorship required to work in the US
Work Experience
Computer Vision Engineer
Octobot
-
Seoul
March 2022
to Present
Completing 4 projects from beginning till the end.
Completed project names: Traffic sign recognition and find location, Car Dashboard analytics,
Anonymization, Drowsiness detection.
Description about every project
Traffic sign recognition and find location - in this project 145 types of Korean traffic signs should be
detected, segmented and every text written inside it should be recognized, should calculate its location
using gps data and deploy to edge devices.
Car
Dashboard Analytics - ADAS, LKAS, speed, time, location and write it into csv file.
Anonymization - blur faces of people and license plate (personal informations) for various purposes.
Drowsiness detection - collect data about driver whether driver is yawning, eyes are closed, which side
driver is looking at.
The most complicated project was traffic sign recognition and find location. The most complicated part
was deploying ready project into edge device. We tried several devices as Jetson nano, raspberry pi 4,
jetson AGX using cameras with VPU chip like OAK-D, OAK-D CM4, ZED2. And we used Jetson AGX and
Zed2 camera.
You may have question like why you chose those devices: The answer would be We tried raspberry pi
and Jetson nano with OAK-D camera. OAK-D's computational power is not enough for model that we
trained. To deploy on that AI camera we should optimize the model and model accuracy may decrease
after optimization. Tried to use OAK-D as stereo camera not use its VPU for model inference but there is
no information to use it as stereo camera. So we decided to use ZED2 camera to use as Stereo camera
and now we need to find device where computation power would be enough for model inference. So we
decided to use Jetson Xavier AGX.
Computer Vision Intern
ITMED Uzbekistan
-
Tashkent
July 2021 to February 2022
Convert Dicom image type to png and helping doctors to label breast cancer dicom images.
Using yolov5 for breast cancer detection in BIRADS system.
Masked-RCNN for breast cancer segmentation
Resnet-50 to classify whether given image is malignant or benign for COVID-19 classification.
Tried to connect Our model to PACS system but failed.
Computer Vision Intern
TASS Vision
-
Tashkent
April 2021 to July 2021
Gathering and preprocessing dataset for the project named Visitor analytics. This helped to increase
model accuracy (100%) in real life tests.
Learning how camera works and testing cameras for spoof attacks for face recognition cameras. Found
several conditions where face recognition is not working.
Education
Bachelor's degree in Computer Engineering
Tashkent University of Information Technologies named after Muhammad al-Khwarizmi - 108 Amir
Temur avenue Tashkent, 100084
September 2022 to Present
Bachelor's degree in Computer Engineering
Inha University in Tashkent - 9 Ziyolilar St, Tashkent
September 2020 to May 2021
College Degree in Taxation
Tashkent Tax College - 3
Little Ring Road, Tashkent 100173
September 2016 to May 2020
Skills
• Linux
• AI
• PyTorch
• Deep learning
• Python
• optical character recognition
• YOLO
Languages
• Uzbek - Fluent
• English
- Intermediate
• Russian - Intermediate
Links
https://github.com/ShakhzodbekYuldoshov
https://www.linkedin.com/in/shakhzodbek-yuldoshov-4ba477203/