Monib Sediqi

Skills

AI Model Development | Python • Pytorch • Tensorflow • Numpy • Scikit-Learn • Seaborn • Matplotlib

Model Orchestration | C++ (Windows & Linux) • AWS Cloud (SageMaker, EC2, S3, IAM, Lambda Function and EFS) • Android • Docker • Kubernetes

Software Engineering | C/C++ Embedded Systems • Java and Kotlin • Android Mobile Engineer • Design Patterns • SQL & MongoDB

Miscellaneous | Computer Vision • Large Language Models (LLM) • FastAPI • Django • Flask

Soft Skills | Team Player • Result Oriented • Organization • Strategize and Execution

Work & Research Experience

Medial IP, South Korea | C/C++, Python, Pytorch, Machine Learning, AWS, Docker, Jira
AI Research Engineer | Nov 2021 - Sep 2023
Idiated and developed algorithms for interactive segmentation and object tracking utilizing prompt engineering techniques including text prompt, bounding box prompt and more.
Developed novel deep learning models based on multimodel engineering for image registration. (Registed a global patent on this)
Deployed ANOVA and Chi-squared testing to select the best possible feature from data for machine learning models.
Designed and developed AI model production pipeline/API to efficinetly ship models from research to production.
Contributed in the deployment of AI models on AWS cloud using SageMaker, EC2, S3, IAM, Lambda Function and EFS

Mayfarm Soft, South Korea | C/C++, Python, Pytorch, Mediapipe, Deep learning, Android SDK
AI Lead Engineer | Jan 2021 - Nov 2021
Successfully led and delivered two AI projects from inception to deployment 1) Gaze estimation, 2) Emotion recognition
Contributed to the development of 2D and 3D gaze estimation algorithms, utilizing techniques such as eye tracking, pupil detection, and calibration.
Conducted extensive experimentation with different machine learning models such as XGBoost, EfficientNet, MobileNet, DBSCAN to solving iris detection problem.
Established the first Korean Eye Gaze Dataset by collecting 120,000 eye images from 60+ participants, showcasing proactive project planning and execution

ZTE Corporation China | C/C++, Design Patterns, Linux, Embedded System, OpenCV
Software Engineer | Jan 2016 - Aug 2017
Developed device drivers for Linux kernels using C/C++
Developed image compression and transmission algorithms for TCP/IP protocols

Education

Jeonbuk National University (JBNU)
Masters in Computer Science & Engineering | CGPA: 3.88/4.0 | Sep. 2018 - Feb. 2021
Courses: Image processing (A+), Machine learning(A), Deep Learning(A) Advanced algorithms & data structure(B+), Advanced Neural Network(A), HPC and parallel computing(A+), Technical writing(A+)

Kabul University, (KBL)
Bachelor of Science in Computer Science with honor | GPA: 92.4% | Mar. 2012 - Dec. 2015
Course highlights: Computer Architecture(A+), Operating systems(A+), Microcontroller & Microprocessor(A), Linux/Unix(A), Programming in C/C++(A+), Java(A+), SQL database(A+), Web design(A+), Data structure(A+), Data communication(A+), Computer networks(A+), Network security(A+)

Said Noor M. Shah High School
High School Education - graduated with honor| Mar. 2009 - Dec 2011

Selected Projects

Foundation Models

Currently invovled in researching one-shot and zero-shot learning for foundation models to add controlability on the output of these models.
Exploring the following conditioning information:

Text Prompt • Image Prompt

Bounding Box • Body Pose

Semantic Map • Depth Map

Inset Image • Style Image

[On Going Project]

Automated Radiology Report Generation (ARRG)

ARRG is the task of generating radiologist like report for x-ray images. In this research I am exploring the usability of prior informaiton such as prior x-ray images and meta information for enhanced automated radiology report generation.

[On Going Project]

C++ TATA API for AI model Deployment

TATA (TisepX Advanced Translation and Analysis) API is used to deploy trained machine learning and deep learning model into windows/linux client applications. I employed data structures best practices for storing and accessing data as well as proposed novel UX design by accompnaying alpha channel with the segmentation mask for better visualization of the results.
I also contributed in deploying of AI model on cloud using Amazon EC2 and integrating it with the backend Fast API.

[More Details] [Demo] [Code]

AI OneClick WholeBody Organ Segmentation from CT Scans

In this project, I was in charge of extending the classes of body organ segmentation and integrating deep learning based segmentation. I deployed vision transformers (ViT) to extend the segmentation capabilities of the software and intergrated MONAI libarary alongside unsupervised learning to enable interactive segmentation.

[More Details] [Demo] [Confidentiality Required]

Lession Detection in Chest X-ray Images

Employed advanced techniques to accurately detect lesion regions in chest X-rays using unlabeled data.
Conceptualization and implementation of an AI model, integrating parameter back projection via GradCAM methodology, to enhance diagnostic accuracy and efficiency.

[More Details] [Code]

AI Virtual Human Twin

Research and development of algorithms for digital twins of human using medical images. (Registed a patnet on this)
Developed a sophisticated deformable/learnable head template that can fit on any 2D image to create 3D image of it.
Photo-realistic rendering of face and whole-body including rendering on NVIDIA omniverse
Ideated and developed a 2D texture mapping deep learning model that maps image texture on a 3D deformable object.

[More Details] [Demo] [Code]

AI 3D Image Registration

Research, development, and deployment of advanced AI applications for medical systems
Developing 3D medical image registration algorithms using deep learning
Initiating the Grand Segmentation Project for medical image segmentation (Pthon and C++)
Applying computer vision algorithm in real-time system software using (C/C++)

[More Details] [Code]

AI Gaze Estimation System

Plan, design, research and development of a comprehensive gaze estimation system software and android app. (Kotlin, Python and C++).

The gaze app achieves an average error of less than 3cm of L2 (Euclidean) distance on android devices - with a screen size of 10.1", and PCs - with a screen size of 24".

[Report] [Demo] [Code]

Smart Scale - AI Pig Weight Estimation

Built, designed and delivered a sophisticated smart scale android app using ML algorithms, point cloud, OpenNI and OpenGL libraries.
The app can estimate and visualize pig's weight from a 3D scanned image. (Java)

The app achieves a 97% of accuracy in real world deployment in less than 7 seconds.

[Report] [Demo] [Code]

CelebrityNet: Korean Celebrity Face Recognition

Design, data modeling and implementation of deep learning algorithm to recognize Korean celebrity's face from an image

[Poster] [Report] [Code]

Deep Learning Based Image Semantic Segmentation

Published in IEEE, CSCI-2020 international conference. In this paper I propose a novel data augmentation strategy that expands the size of the dataset to improve the performance of deep learning networks for image semantic segmentation.

[Paper] [Demo] [Code]

3G Mobile Service App

Design and devlopment of 3G mobile service app to fetch information from telecom operators and provide them to the customers through easy, and user-friendly interface

[Poster] [Report] [Code]

Patents

Medical Image Registration Method and Apparatus, PATENT-2022-0132732 (Registered)

Digital twin of the face and whole body from CT images using a learnable 3D template model (Under Review)

Publications

A Novel Upsampling and Context Convolution for Image Semantic Segmentation | Journal of Sensors

Sediqi KM, Lee HJ. A Novel Upsampling and Context Convolution for Image Semantic Segmentation. Sensors. 2021; 21(6):2170. https://doi.org/10.3390/s21062170

Improved Image Semantic Segmentation Based on Cascade Data Augmentation | Las Vegas, USA
Sediqi KM, Lee HJ. Improved Image Semantic Segmentation Based on Cascade Data Augmentation. In 2020 International Conference on Computational Science and Computational Intelligence (CSCI) 2020 Dec 16 (pp. 1607-1612). IEEE.

Importance of context in semantic segmentation of images | Jul-2020, Busan, South Korea
Khwaja Monib Sediqi,and Hyo Jong Lee. "Importance of context in semantic segmentation of images." 한국정보과학회 학술발표논문집 2020.7 (2020): 871-873.

Extended Temporal Convolutions for Human Action Recognition in Videos| Tianjin, China
Sediqi KM, Sung TL, Lee HJ. "Extended Temporal Convolutions for Human Action Recognition in Videos." International Symposium on Information Technology Convergence (2019)

Decomposed "Spatial and Temporal" Convolution for Human Action Recognition in Videos | Seoul, South Korea
Sediqi KM, Lee HJ. "Decomposed "Spatial and Temporal" Convolution for Human Action Recognition in Videos."한국정보처리학회 2019년도 춘계학술발표대회 (2020)