Project
I have listed some of my projects here. You can find the source code by following the github link under each project section. The details buttons will show some snapshots of the project in action or in some cases the results that had been achieved. If there is a live button, then that project is available online. If you hover over live button, it will show the platform where it is been hosted.
LabelAug
labelaug is a simple python GUI app that is written in python and uses QT for graphical interface. It allows user to easily augment images and their corresponding labels in YOLO format. It offers multiple basic augmentation techniques which are most commonly used for data augmentation. Users can also see the bounding boxes on their images as well as export the image as a .jpg file. It takes care of augmentation of multiple images and labels in a given directory without any sweat! For large number of files and images, the augmentation process works in different thread which offers users an uninterrupted interaction with the GUI.
Breast Cancer Classification
Transformed four ML algorithms such as SVM, DTC, KNN, RF to develop a Hard Voting meta-classifier, which classified tumor with 99.42% accuracy. The Wisconsin Diagnostic Breast Cancer (WDBC) dataset was used for this work and published in the 11th International Conference on Electrical and Computer Engineering (ICECE) 2020. This was a group project.
PointExtract
PointExtract is a simple python GUI app that is written in python and uses QT for graphical interface. The motivation story behind this tool stemmed from the necessity of extracting points from a constellation image to accurately visualize in a CLI tool. That means getting the pixel co-ordinates of some points of an image and plot it with the same aspect-ratio. This tool also offers 0-360 degree rotation. So, any set of points can be rotated with respect to their mean position. The user can also save the coordinates as a list in .txt format to their local space for their need.
RedPortalGreenPortal
Inspired by the game Snakes and Ladders, I have develop this game from scratch where there is a fixed number of tiles that a player has to cross before others to win. Instead of snakes and ladders this game has multiple portals of two kinds. Implemented player select options allowing users to play single player mode against computer and 2 player mode. Users have also options to choose board size. The whole game is developed using only Javascript, HTML and CSS.
Empty Space Detection in Retail Environment
This project aims to detect empty spaces in retail environment like Walmart, HEB or any other supershops. A smaller subset of SKU110K dataset has been used. The original data labels are for the purpose of detecting products. Only images are used from the original dataset and the empty spaces are labelled to generate the annotations files. For labelling, LabelImg was used and all the labels were created in YOLO format initially. Multiple small python sripts were developed to preprocess images and augmentation. Also all labelled .txt files were converted into a single .csv file which includes all necessary label information that can be used for getting key information about training data and training different models like Faster R-CNN, YOLO or SSD. A precision of 75.7% has been achieved by training a YOLO v.7 model on Google Colab so far. This project certainly has future scopes for further developments.
Stars of our sky
This web app's objective lies in creating a simple informative platform from which users can easily access information about our sky's STARS and which Constellation they belong to. There are plenty of resources out there that have detailed information about stars, planets, solar systems, constellations, and a lot more. The first problem is to collect all the constellation's names. Luckily a .csv file was found from Kaggle's dataset. This dataset also includes the English name beside the Latin name of constellations as well as the brightest star of that constellation.
For scrapping the information from Wikipedia it was noticed that their URLs have specific format for single-word constellation names and double-word constellation names. BeautifulSoup was used to scrape the information. Prior to scrap information from Wikipedia, our dataset was processed accordingly, so that http request does not end up in a 404. Then the information was saved as js objects for faster loading of the web app. The image URLs were also saved as in the file. Showing a constellation's pic from an API call was tried but failed due to inappropriate tagging of API images or multiple meanings and various usage of constellation names in multiple situations/objects. An API from API Ninja called 'Stars' was used to display more information for each constellation's brightest stars.
