Hi, there 
I am
Bekhzod Olimov
A Senior AI Engineer & Researcher
with 5+ years of experience in ML/DL
Info
Hi! I am Bekhzod, an experienced AI/ML engineer with a PhD degree in Computer Science & Engineering. My passion is to apply AI techniques in various applications to make people's life easier and more comfortable.
Here are my main skills:
Projects
Image classification task-related research
I have conducted extensive research and completed several projects focused on deep learning (DL)-based image classification, particularly in the areas of activation functions and model architecture. The outcomes of my research have been published in international SCIE journals, such as Nature Scientific Reports and Concurrency and Computation.
Semantic Segmentation task-related research
I have undertaken in-depth research and completed several projects focused on deep learning (DL)-based semantic segmentation, particularly in the fields of autonomous driving and medical imaging. The outcomes of my research have been published in international SCIE journals, such as IEEE Access and Multimedia Systems.
Object detection task-related research
I provided instruction to undergraduate students on conducting deep learning research, writing a scientific paper, and presenting it at a scientific conference. The research focused on detecting signs of disabled individuals in vehicles. The project was initiated from scratch, involving comprehensive tasks such as data collection, model formulation, and inference.
Background Removal Project
I conducted on background removal using the latest deep learning-based binary semantic segmentation techniques.
Anomaly detection-related research
I conceived an idea, wrote a proposal, and secured a National Research Fund of Korea grant for a three-year research period. The results of this research were published in SCIE journals, including "Computers & Industrial Engineering" with an impact factor of 6.7, and "Nature Scientific Reports".
Crowd Counting-related research
Following the Seoul Halloween crowd crush in October 2022, we conducted research to identify densely populated areas and preemptively alert law enforcement about potential accidents using Deep Learning techniques.
Face Verification using AI Project
I developed a face verification system that can be used to check the attendance of students or employees at school or office. I used traditional and DL-based computer vision techniques during the project. This project is integrated to the various organizations in Uzbekistan.
3D Medical Image Segmentation task-related research
I conducted research on segmenting 3D medical image data (specifically, body organs, like spleen, prostate and so on along with dental nerve. The latter was a joint project with a South Korean dental clinic. The segmentation AI model is integrated to the South Korean dental clinic software.
Image Retrieval task-related research
I undertook in-depth research on developing a model for retrieving real-life images based on sketch images. The trained model was implemented to detect counterfeit luxury goods in the Customs office of the Republic of Korea and to detect used car parts in the GPARTS web-site.
Kaggle Projects using open source databases
I engaged in over 200 mini projects using open-source databases on Kaggle, focusing on various AI tasks including image classification, semantic segmentation, object detection, natural language processing, and more.
Manga2Webtoon using AI
Embarking on a project to convert Japanese and Chinese manga into Korean Webtoons using AI involves several sequential steps. These include segmenting parts using semantic segmentation, applying color to the original grayscale images using image generation models, extracting text from speech bubbles using Optical Character Recognition (OCR), and more.
Cervical Cytology Screening & Early Detection of Cervical Cancer using CAD
Optimizing deep learning-based AI models for early detection and classification of abnormal cells, improving diagnostic accuracy, and reducing false positives. This project aims to enhance the efficiency of cervical cancer screening processes, ultimately supporting early intervention and better patient outcomes.
Multimodal Medical Image Classification
Dual input (of an image and text) allows for more accurate diagnosis and decision-making in healthcare, especially in scenarios where visual data alone may be ambiguous, leading to improved clinical insights and personalized treatments for patients.
Remove Anything with Multiple Points
This mini-project leverages the Segment Anything model for semantic segmentation to identify and isolate objects in images. It then uses the LaMA 2 inpainting model to seamlessly remove the selected objects. Unlike the original Remove Anything model, which does not support multiple segmentation points, this project addresses that limitation by enabling segmentation of multiple objects simultaneously, enhancing the flexibility and accuracy of object removal for complex tasks.
Multi Instance Learning
This project implements a robust Multi-Instance Learning (MIL) framework designed to classify complex data where individual instances belong to different categories within a larger bag. The system dynamically learns bag-level labels from instance-level features, leveraging advanced AI techniques for both interpretability and performance. By optimizing attention-based mechanisms, the model identifies and weighs critical instances, making it highly effective for challenging tasks like image and text fusion in multimodal datasets.
Cell Detection Project using YOLOX
This project leverages the advanced YOLOX object detection framework for precise cervical cancer cell detection in medical images. The system is designed to identify and localize cells with high accuracy, even in challenging datasets. By integrating custom data augmentation, anchor-free detection, and dynamic label assignment techniques, this project ensures robust performance and scalability for medical imaging applications. Additionally, inference speed is significantly enhanced by utilizing TensorRT, optimizing the model for faster, real-time processing on NVIDIA GPUs, making it ideal for clinical and large-scale deployments.
Licence Plate Detector from Image and Video
Developed a License Plate Detector that accurately identifies license plates in both images and videos using a combination of object detection models and image processing techniques. The project includes real-time video frame analysis, bounding box visualization, and an interactive UI for uploading and testing various media formats.
Contact
Do you have any AI/ML-related application in your mind?
Contact me now for a FREE consultation.
