(May 2024 - Present)
Software Engineer
Teton Private Ltd.
constcoder={name:'Fahad Mohammad Rejwanul Islam',skills:['Python', 'Computer Vision', 'OpenCV', 'MMCV', 'Image Processing', 'TensorFlow', 'Torch', 'CNN', 'YOLO', 'LSTM', 'MySQL', 'MongoDB', 'NestJS', 'FastAPI', 'NodeJS', 'JavaScript', 'ReactJS', 'ExpressJS', 'AWS', 'TypeScript'],hardWorker:true,quickLearner:true,problemSolver:true,hireable:function() {return(this.hardWorker&&this.problemSolver&&this.skills.length>=5);};};ABOUT ME
Who I am?
My name is FAHAD MOHAMMAD REJWANUL ISLAM. I am a professional and enthusiastic programmer in my daily life. I am a quick learner with a self-learning attitude. I love to learn and explore new technologies and i am passionate about problem-solving. I love to dive in computer vision, image processing, MLops and web application problems. I am available for any kind of job opportunity that suits my skills and interests.
Experiences
(Dec 2022 - Dec 2023)
Teaching Assistant(ST)
BRAC University
(Jan 2021 - Present)
Self Employed
Code and build something in everyday.
Skills
PROJECTS
Leveraging Sequential Deep Learning Models for Detecting Multitude of Human Action Categories
constproject={name:'Leveraging Sequential Deep Learning Models for Detecting Multitude of Human Action Categories ',tools: ['Python', 'Tensorflow', 'Torch', 'YOLO v8', 'CNN', 'OpenCV', 'LSTM', 'CONV-LSTM', 'LRCN],myRole:Image processing, Model training, Result analysis,Description: Human Action Recognition (HAR) is critical for smart decision-making and public safety. Our study focuses on recognizing actions using a dataset of 1,275 videos featuring 20 actions, including violent and non-violent behaviors. A pipeline combining YOLO-v8 for background extraction and deep learning models (Conv-LSTM and LRCN) was developed. The LRCN model achieved 62% accuracy and 60% F1 score for 20 classes, 63% accuracy and 66% F1 score for 17 classes, and 88% accuracy with 87% F1 score for binary classification. This highlights HAR's potential for enhancing real-time safety systems.,};Cattle-Biometric Identification System
constproject={name:'Cattle-Biometric Identification System',tools: ['Python', 'OpenCV', 'HogDescriptor', 'Tensorflow', 'Torch', 'Numpy', 'MySQL', 'FastAPI],myRole:Model training and integrating the model in the api for muzzle identification,Description: This is the biometric identification system of a cow, where you can register muzzle images of cow and verify through the muzzle image , is it registered or not. If the muzzle is already registered then it will show the muzzle is matched or else not matched.,};Cattle-Weight Prediction System
constproject={name:'Cattle-Weight Prediction System',tools: ['Python', 'MMCV', 'Joblib', 'Matplotlib', 'Keras', 'Torch', 'Tensorflow', 'Pillow', 'PyYAML', 'Scikit-learn', 'Seaborn', 'SQLAlchemy', 'Torchvision', 'vine', 'xtcocotools', 'MySQL', 'FastAPI],myRole:Image Processing, model training and integrating the models in the restful APIS for cow weight prediction,Description: This system uses side and rear images of a cow to predict its weight. It segments the images, detects key points, and calculates the cow's length and girth. A regression model then predicts the weight based on these measurements. Multiple models are used for segmentation, keypoint detection, and regression, with MMLab as a key tool for implementation.,};Advanced Leaf Counting for Agricultural Applications Using Deep Learning
constproject={name:'Advanced Leaf Counting for Agricultural Applications Using Deep Learning',tools: ['python', 'torch', 'CNN', 'DE-CNN', 'Torch', 'Tensorflow],myRole:Model training,Description: This project focuses on counting rosette leaves in RGB images for plant phenotyping using adaptable data-driven methods. A deconvolutional neural network (De-CNN) handles segmentation, while a CNN counts the leaves. Despite limited training data from the CVPPP-2017 dataset, we achieve strong results with data augmentation, achieving an average mean absolute count difference of 1.62 and a standard deviation of 2.30, outperforming previous methods.,};User and Admin Management Service
constproject={name:'User and Admin Management Service',tools: ['NestJS', 'ReactJS', 'Redis', 'Prisma', 'Javascript', 'Typescript', 'AGORA],myRole: Full-stack development,Description: This project delivers a seamless platform for users and administrators. Users can register, log in, access real-time chat via WebSocket, and enjoy live streaming with Agora. Administrators manage the system with tools for job allocation, salary settings, and user account management. Built with Vite and NestJS, it ensures a fast, scalable, and efficient experience,};Integrated Surveillance System for Threat Detection
constproject={name:'Integrated Surveillance System for Threat Detection',tools: ['Python', 'OpenCV', 'TensorFlow', '3D CNN', 'Torch', 'Keras', 'YOLOv8', 'InsightFace', 'NumPy', 'Farneback Optical Flow],myRole:Implemented and integrated deep learning models for violence detection, weapon detection, and face identification.,Description: A real-time surveillance system that detects violence, weapons, and identifies faces in video streams.The system outputs annotated videos for use in security and law enforcement monitoring.,};Educations
2020 - 2024
Bachelor of Science in Computer Science and Engineering (B.Sc. in CSE)
Major: Computer Science and Engineering
CGPA: 3.81
BRAC University
2017 - 2019
Higher Secondary Certificate (HSC)
Major: Science
Dhaka Residential Model College
2015 - 2017
Secondary School Certificate (SSC)
Major: Science
Naogaon Govt K.D High School
PUBLICATIONS
Leveraging Sequential Deep Learning Models for Detecting Multitude of Human Action Categories
Authors - [Kazi Al Refat Pranta; Fahad Mohammad Rejwanul Islam; Khandakar Fahim Ahmed; Prince Saha; Naimur Rahman]
Description - Multiple person's action detection from video data.
Date of Publication - December 19, 2024