I am a Ph.D. Research Scholar working in Computer Vision. I am passionate about learning machine learning algorithms and solving practical challenges in their application to real-world problems.
My research concentrates on the use of machine learning (especially Deep Learning) methods for activity recognition from videos. A deep learning architecture is proposed for egocentric video, which integrates both action appearance and motion within a single model. Development and evaluation will be based on publicly available data.
I'm interested in computer vision, deep learning, machine learning, image processing and video analysis.
Human Activity Recognition, On going, 2024
Publication
Will be updated.
Sheep Lambing Detection, On going, 2024
Publication
Will be updated.
Noise Profiling for ANNs: A Bio-inspired ApproachOn going, 2024
Publication
A novel approach to noise profiling for artificial neural networks (ANNs) is proposed, which is inspired by the sensory systems of insects. This approach entails the utilization of both Gaussian and Chaotic noises to enhance the adaptability, learning and generalization capabilities of ANNs.
Virtual laboratories are an essential part of E-learning because all the students in their institutes may not have sufficient lab facilities. These experiments can be accessed from anywhere and anytime. Therefore, the Ministry of Human Resource Development (MHRD), Govt. of India took an initiative of integration of virtual laboratories under the national mission on Education through Information and Communication Technology (NME-ICT) . The motive of virtual lab integration is to make all the developed projects into an open source repository such that all the lab information is available to a community, students as well as academic institutes, for use and development, to convert all licensed contents into a platform that is independent of any licensed software.
The Ministry of Human Resource Development (MHRD), Govt. of India took the initiative of Remote Triggered Digital System Laboratory under the National Mission on Education through Information and Communication Technology (NME-ICT). This virtual laboratory provides the theoretical understanding of digital electronics to the students by performing various experiments.
The Human Management System is a Java based (J2EE) system which provides intranet automation of HR software. The aim of the paper is based on a project that helps the overall management of the employees, who work in a company. The proposed system contains all the information regarding employees in the company. The system is developed on good interaction as well as communication facilities between the HR administrator and the working employees.
Attented Conferences and Workshops
The AI Research Hub Symposium, Aberystwyth University, United Kingdom, September 2023.
The 22nd UK Workshop on Computational Intelligence, Aston University, United Kingdom, September 2023.
Faculty of Business and Physical Sciences Postgraduate Research Conference 2023, Aberystwyth University, United Kingdom, July 2023.
1st AI Summer School for Beginners, Aberystwyth University, United Kingdom, August 2022.
5th Summer school on Artificial Intelligence, Indian Institute of Information Technology Hyderabad, India, August 2021.
IEEE 5th International conference for Convergence in technology, Pune, India, March 2019.
IEEE Fourteenth International Conference on Information Processing (IcInPro), Bangalore, December 2018.
IEEE 3rd International Conference for Convergence in Technology (I2CT), Pune, India, April 2018.
MATLAB workshop in Reflux 7.0, Indian Institute of Technology, Guwahati, India, 2019.
Workshop in Data Science in Financial technology, Research Conclave, Indian Institute of Technology, Guwahati, India, 2019.
Workshop in machine learning, Research Conclave, Indian Institute of Technology, Guwahati, India, 2019.
Poster presentation in Research Conclave, Indian Institute of Technology, Guwahati, India, 2017.
A one-day workshop on virtual laboratory jointly organized by Assam Engineering College and Indian Institute of Technology, Guwahati, India, February 2017.
A One-day workshop on virtual laboratory jointly organized by NIT Meghalaya, Shillong and Indian Institute of Technology, Guwahati, India, November 2016.
A One-day workshop virtual laboratory jointly organized by Central Institute of Technology, Kokrajhar and Indian Institute of Technology, Guwahati, India, August 2016.
Virtual Labs Summer Sprint Integration workshop, Indian Institute of Technology, Guwahati, India, 2015.
First Integration workshop, International Institute of Information Technology Hyderabad, India, 2014.
Attended as a volunteer, 33rd Foundations of Software Technology and Theoretical Computer Science, Indian Institute of Technology, Guwahati, India, 2013.
Fun Projects
Following are a collection of practice projects, which sparks my interest in further explorations.
In this project, the goal was to train a Convolutional Neural Network (CNN) for regression prediction with Keras and then train a CNN to predict house prices from a set of images.
The objective of the project was to create a deep learning model to classify images of clothing from the Fashion MNIST dataset. The Fashion MNIST dataset is a collection of grayscale images of 10 different categories of clothing and accessories, like T-shirts, trousers, pullovers, dresses, coats, sandals, shirts, sneakers, bags, and ankle boots.
Smile detection with OpenCV, Keras, and TensorFlow Github
This project used Haar cascade face detector, extract the face region of interest (ROI) from the image and then pass the ROI through LeNet for smile detection.
Breaking captchas with deep learning, Keras, and TensorFlow Github
This project demonstrated how to use deep learning techniques, specifically with frameworks like Keras and TensorFlow, to automatically solve CAPTCHA challenges.
Use Checkpoint Strategies with Keras and TensorFlow Github
This amied to use Early Stopping and Model Checkpointing in training Keras models encapsulates a sophisticated approach to deep learning.
ImageNet: VGGNet, ResNet, Inception, and Xception with Keras Github
The project was designed to classify an image by identifying the main subject in the image, leveraging pre-trained deep learning models available through TensorFlow's Keras library. It accepts an image file and a model name as input parameters. The script supports various state-of-the-art image classification models like VGG16, VGG19, ResNet50, InceptionV3, and Xception, which have been trained on the ImageNet dataset.
The goal was for building, training, evaluating, and plotting the performance of a convolutional neural network (LeNet) for digit classification on the MNIST dataset.
The goal was to create the first deep learning neural network model in Python using Keras. Here, we started by loading and preparing our dataset, followed by defining and compiling a Keras neural network model. We trained the model on our data, evaluate its performance, and then use it to make predictions on new data. We used Pima Indians onset of diabetes dataset.
We explored hands-on code that illustrates how to implement and apply convolution operations and kernels to images. This insight aided in understanding the internal workings of Convolutional Neural Networks (CNNs) during their training phase.
The project demonstrated how a perceptron model could learn bitwise operations through a basic machine learning process involving training with input features and corresponding labels, followed by testing to evaluate the model's predictions.
Pedestrian Detection with 4 Different Computer Vision Techniques Github
This project explores pedestrian detection using four different computer vision techniques.
Method 1: Background Subtraction + Contour Extraction
Method 2: Haar Cascades (Viola-Jones Classifiers)
Method 3: Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM)
Method 4: Single Shot Detector (SSD) with MobileNet
This project showcased the implementation of Haar Cascade classifiers for object detection in video streams. Haar Cascades are a popular method for object detection due to their efficiency and effectiveness, particularly in detecting faces and other predefined objects. Using OpenCV, this project demonstrates how to apply Haar Cascades to real-time video data to identify and track objects.
The Smile Detection Project aimed at identifying smiles real-time video feeds using a facial landmark detector to accurately determine the presence of a smile.
These project series provides an essential overview of computer vision techniques using OpenCV. It begins with fundamental image operations—loading, displaying, and pixel manipulation—then advances to drawing, translation, rotation, resizing, flipping, and cropping. Additionally, it explores arithmetic operations, bitwise manipulations, masking, and channel manipulation. Accompanied by downloadable source code for each tutorial, this series offers a practical and efficient way to grasp the key functionalities of OpenCV, making it perfect for beginners eager to learn quickly.