Kesha Bodawala

The aim of this project was to integrate Tensorflow's object detection API with OpenCV and OpenGL to code 2D and 3D augmented reality (AR) applications. With the help of Tensorflow's API, I trained MobileNet V1 neural network architecture to detect (1) my drawing and (2) any smartphone. These detected objects were used as markers for the applications. A 2D video was augmented upon detection of my drawing using OpenCV and a 3D animation was augmented upon detection of any smartphone using OpenGL.

This project involved designing a deep learning algorithm based on LSTM networks for stance detection as the first step to combat fake news problem. The model used two LSTM encoders, a simple encoder to encode headline and a hierarchical LSTM network encoder to encode text, and used a feed forward neural network to compare the encodings predict the result. The project was coded in Python using Keras. GPU provided by Google Colab was used for model training. I experimented with a variety or… Show more

This project involved designing a deep learning algorithm based on LSTM networks for stance detection as the first step to combat fake news problem. The model used two LSTM encoders, a simple encoder to encode headline and a hierarchical LSTM network encoder to encode text, and used a feed forward neural network to compare the encodings predict the result. The project was coded in Python using Keras. GPU provided by Google Colab was used for model training. I experimented with a variety or pre-processing steps, batch processing to minimize padding and a couple of learning rate schedules with warm restart such as cosine annealing. Show less

In this project, I learned concepts of Recommender Systems by. I studied and compared various machine learning based solution. I received appreciation by my professor for excellent understanding of the concept and well-organized summarization of the final project report.

In this project, I engineered a method to detect Facial Expressions OpenCV library in Python. This utilized a feature extraction method similar to Histogram of Oriented Gradients (HOG), and k-Nearest Neighbour classifier with Chi-square distance to predict facial expression. I was able to achieve 98.24% accuracy on JAFFE dataset.

This project was developed using MATLAB. It was a novel approach to generate superpixels using Breadth First Search Technique and the state-of-the-art SLIC (Simple Linear Iterative Algorithm). This technique generates superpixels whose boundary adherence exceeds SLIC's boundary adherence by maximum 26.59%.

I did this project under Prof. Kishor Upla. A Virtual Newspaper, just like a real newspaper, was design that reveals new articles as the parts of foldable display (virtual newspaper) are exposed to camera. I implemented concepts of thresholding, contour detection, pattern recognition and perspective transforms using OpenCV libraries in C on Microsoft Visual Studio IDE.

This project was done as a part of my internship at Indroyd Labs. I used depth sensor of Microsoft Kinect Skeleton D2D SDK, and developed C++ based code for an application to control powerpoint presentation using hand gestures. This project was used as a baseline application for various sales pitches.

This project was done under Prof. Anand Darji. The developed system detected eye winks by analyzing frequency and power of signals captured from F7 and F8 channels of electroencephalogram (EEG).