Fall 2018
Name: PATRON: Perpetual Abnormality Tracking with Robust Ochlocracy Neutralization
Course: Directed Research (CSE498R)
Supervisor: Mr. Adnan Firoze
Language(s) and Framework(s): Matlab, Python, C++
Summary: Machine involvement in analyzing video data is a demand of time. Bangladesh, a country with emerging potentiality does not have any machine involved video understanding context, rather than inefficient manual intervention. However, several political and criminal activities are regular in this country where immediate surveillance could instantly conduct standard protection against this ochlocratic occurrence. As per the solution, we propose a state-of-the-art technique for abnormality tracking from video data in an efficient manner, along with machine-involved faster ochlocracy neutralization.
Name: QUICKML: Qualitative User Interface For Competitive and Kinematic Machine Learning Experiments
Course: Software Engineering (CSE327)
Instructor: Dr. Mohammad Ashrafuzzaman Khan
Language(s) and Framework(s): Python, Flask, SQLite, Numpy, Scipy, Pandas, Matplotlib, jQuery
Summary: Machine learning has been an emerging part in most domains requiring statistical analysis over data. Regarding the content, learning machine learning became an important part of being specialized in the data analysis extent. Due to the expertise in the field, an easy way of learning machine learning is one of the most demanded factors for all people. As per being in still research field, a specific tool for facilitating the whole experiments for several people has been poorly designed. In this project, we are aiming to overcome the limitation with a user-friendly Web GUI which allow users to test their skills with the whole standard machine learning standard experimental steps. Also, we intend to design a contest-based machine learning system for comparing output result with the own-uploaded dataset and improve with further parameter regularization.
Name: Design and Implementation of a WiFi Controlled Scrolling LED Display
Course: Operating Systems Design (CSE323)
Instructor: Dr. Saeed Mahmud Ullah
Language(s) and Framework(s): Arduino C, ESP8266 (NodeMCU)
Summary: In this project, we are aimed to demonstrate a basic display text with WiFi control function. The whole pipeline would consist on creation of an wifi hotspot signal, later creating an web server as display interpreter through web application, later mapping the filtered input text from user to the matrix panel and do continuous text movement in the aligned display matrix. In the sections below, we will demonstrate the challenges, components and total experimental pipeline along with the coding and hardware-software interaction with the microcontroller development kit involved in the whole extent.
Summer 2018
Name: Oboyob: A Sequential-Semantic Bengali Image Captioning Engine (demo/code available upon request)
Course: Neural Networks (CSE448)
Instructor: Dr. Mohammad Rashedur Rahman
Language(s) and Framework(s): Python, Matlab, Keras (TensorFlow Backend), Numpy, PHP, MySQL, Javascript
Abstract: Understanding the context with generation of textual description from an input image is an active and challenging research topic in computer vision and natural language processing. However, in the case of Bengali language, the problem is still unexplored. In this paper, we addressed a standard approach for Bengali image caption generation though subsampling the machine translated dataset included with several pre-processing techniques toward involvement of state-of-the-art CNN-LSTM architecture-based models. The experiment was conducted on standard Flickr-8K dataset, along with several modifications conducted to adapt Bengali language. Training caption subsampled dataset was computed for both Bengali and English language for further experimentation with 16 distinct models developed in the entire training process. Further, the trained models for both languages were analyzed with respect to several caption evaluation metrics with judgement of network architecture performance and caption performance trade-off. Further, we established a baseline performance in Bengali image caption context.
Publication: A Journal Paper Under Review
Name: Sentiment Perception and Probabilistic Opinion Inference From Bangla Textual Data (demo/code available upon request)
Course: Machine Learning (CSE445)
Instructor: Mr. Mirza Mohammad Lutfe Elahi & Ms. Silvia Ahmed
Language(s) and Framework(s): Python, Matlab, C++, Java, Numpy, Scikit-Learn, Pandas, Matplotlib, PHP, MySQL
Abstract: Sentiment classification is a way to mine people’s opinions, sentiments, evaluation towards societal entities such as products, events, etc. Understanding accurate, and highly effective sentiments from Bangla Language is a challenging task. In this project, we aimed to classify sentiments of Bangla comments from social media and blog sites. In our process, we collected complete Bangla comments from internet, manually labelled ground truth by contributors. For the experiments, we collected two types of comments, bipolar, and multi-tagged. Later, the collected bipolar comments were pre-processed, followed by classification. The multi-tagged outputs were given input texts were pre-processed in same way, despite being the multi-label data, the output was a regression problem rather than classification problem. Further, we evaluated the bipolar data model with respect to computational cost and accuracy, and the multi-tagged data with respect to lower cost regression output. Later, we examined the proposed data models, obtained state-of-the art accuracy, along with the regression data model.
Spring 2018
Name: Design and Implementation of an 8-bit Single Cycle CPU (demo/code available upon request)
Course: Computer Organization and Architecture (CSE332)
Instructor: Dr. Tanzilur Rahman
Language(s) and Framework(s): Java, Logisim, Swing (Java)
Summary: Our task was to design an 8 bit single-cycle CPU that has separate Data and Instruction Memory. It should also be able connect with at least 2 output devices. We use an accumulator based architecture, where we will write and read data on accumulator. This accumulator will be connected to the register file to facilitate data movement from the registers to the accumulator and vice versa. One Operand. Operands are register based. And saved to or fetched from accumulator $acc. The addressing field is 4-bit long, therefore there can be a maximum of 16 registers. We will do 16 operations based on our developed architecture. These operations will be used for various arithmetic calculations, conditional check, and jump. Furthermore, all the operations are mentioned in “Instructions Table” Section. We chose 3 formats which have been described at “Formats” section. List of the registers are illustrated at “Register Table” section with their naming and functionalities.
Fall 2017
Name: Design and Implementation of Particular Text with Combinational and Sequential Circuit (demo available upon request)
Course: Digital Logic Design (CSE231)
Instructor: Dr. Arshad M Chowdhury
Language(s) and Framework(s): Logisim
Summary: The main idea of the project is project is to deploy a total circuit which will automatically show CSE231-2-7 in single 7-segment display. It can be in both forward or reverse (7-2-132ESC) way. Total design of this circuit will be designed in three phases. In the first phase, we will design the combinational part of the circuit, which means we will specify our inputs and what will be the outputs for the corresponding outputs for the circuit and we will go on formulation step. After formulating them, we will further implement the circuit with our theoretical simplified designed, and finally check and verify if we are getting all the expected outputs for the inputs as per defined in the “specification” step, also our desired one. Later, we will design the sequential circuit with the facilitation of forward and reverse sequence. Further, the sequence will be implemented according to prior state diagram design according to the procedure.
Summer 2017
Name: Efficient Facial Recognition in Constraint Environment with Model Hybridization (demo/code available upon request)
Course: Junior Design (CSE299)
Supervisor: Mr. Adnan Firoze
Language(s) and Framework(s): Python, C++, Lua, OpenCV, Dlib, Scikit-Learn, TensorFlow, Keras
Abstract: In this paper, the main objective is to make face recognition system faster by reducing recognition time without compromising accuracy for a constrained environment i.e. classroom, and provide a comparative review of state-of-the-art and classical approaches considering multiple faces that are at variable distance from the camera in the same image. This makes it a more challenging problem. Several models have been developed to partition the faces from a test image into three different levels. We have developed model hybridization by applying some classical but faster face recognition models namely Eigenfaces, Fisherfaces, Local Binary Patterns (LBP), and state-of-the-art yet relatively slower Convolutional Neural Network Model (CNN). Our proposed model hybridization technique based on different levels done by face partitioning has achieved approximately 33.43% faster performance than CNN while maintaining accuracy same as of CNN of our own dataset of faces of a classroom of 15 students while a class was going. The faces were of different students in different places, positions, poses and lighting. The objective of our research is not to enumerate and show how large a dataset we can identify by face, rather it is more interesting. We are interested in recognizing multiple faces at different distances from camera (hence, varying size, posture etc.) which calls for a unique approach as opposed to large dataset headshots from different angles of single people. We used different classification models for different levels of distance from the camera to achieve this faster response, making it a novel hybrid model.
Publication: Firoze A, Deb T. Face Recognition Time Reduction Based on Partitioned Faces without Compromising Accuracy and a Review of state-of-the-art Face Recognition Approaches. In Proceedings of the 2018 International Conference on Image and Graphics Processing 2018 Feb 24 (pp. 14-21). ACM.
Name: NSU ECE Project Database (demo)
Course: Database Systems (CSE311)
Instructor: Mr. Adnan Firoze
Language(s) and Framework(s): HTML, PHP, Javascript, MySQL, Oracle Database
Summary: This project was built by aiming to create a universal academic project management system. Here, the faculty members can create project information, add students, and upload files for the corresponding projects. From the admin panel, the students are graded for their projects. Each of the projects will have only one supervisor and several students. We implemented several JOIN queries for giving the outputs. In the homepage, some information related to the projects and supervisors is shown (see the demo).
Spring 2017
Name: Advanced Contact Management System (demo/code available upon request)
Course: Data Structures & Algorithms (CSE225)
Instructor: Dr. B. M. Mainul Hossain
Language(s) and Framework(s): C++
Summary: In this project, we focused on the implementation of different data structures to build a complete contact list management system. We have used Linked List, Binary Search Tree, QuickSort, Trie Data Structure to build the overall console based software.
Fall 2016
Name: My Synopsis - A Simple Faculty Profile Builder (demo)
Course: Introduction to Object-Oriented Programming (CSE215)
Instructor: Mr. Adnan Firoze
Language(s) and Framework(s): Java, Swing Framework, JSON, HTML
Summary: This project was focused on generating static profile webpage, especially for the faculty members. This Swing GUI based interface has the support of saving and retrieving data using JSON. The image files are converted into Base64 encoded image for simple HTML rendering. Finally, there was a feature to upload the HTML file into FTP server with given credentials.