Meghana Kumar's repositories
pixel-level-accuracy
Considering an open-source Mask R-CNN implementation, which is a deep multitask learning solution for object detection, instance segmentation and keypoint annotations. The current task involves given an annotated instance segmentation image and the original image, the pixel level accuracy of each instance/class from the image is calculated. This implementation does not contain any fine-tuning to support extra classes than what the model has been trained for initially.
DLGalaxyMorph
Using Deep Learning tools in Galaxy Morphology as part of Inter-University Center for Astronomy and Astrophysics (IUCAA)
multispectral-maskrcnn
The goal of this project is to ascertain the role of intensity and depth channels in instance segmentation of traffic signs. It has been shown that secondary channels, like intensity and depth, provide complementary information to the neural net and can thus boost the efficiency of a model. The project aims to explore the ramifications of experimenting with intensity or reflectivity and depth channels along with RGB images and determine the optimal architecture to combine them most effectively. The intensity and depth data is obtained from LIDAR sensors. Two different implementations of substituting channels in Mask R-CNN are explored.
HierarchicalClustering_PhylogeneticTree
An implementation of Agglomerative and Divisive Hierarchical Clustering on Human DNA Genetic Data in the FASTA format.
algorithms-implementation
Various implemetation of algorithms in Data Structures, Design and Analysis of Algorithms
tweet-clustering-repost
1. Finding the “true” number of posts, by grouping the Twitter posts together if they are retweeting or commenting on the same message. Your method should be robust to small changes, so that if someone introduces a small typo or uses a retweet prefix or postfix that’s not in the set of data provided, your technique for grouping would still work. 2. Ranking these groups of Twitter posts by popularity.
type-check
Type Checking of a c-like language as mini-project in Principles of Programming Language
compilers
An implementation of Scanner, Parser, and Three Address Code Generator for a mini C-like language
deal-sourcing-hackernews
Python scraper to scrape the job postings from Hacker News “Who is hiring” posts and determine relevant information with Named Entity Recognition.
learning-pytorch
Example problems while learning pytorch
opencv_contrib
Repository for OpenCV's extra modules
prob_generative
Probabilistic Generative Model implemented on training andd testing data
sqname-generator
Sentence Tranformers - a state-of-the-art NLP model for detecting semantic similarity is fine-tuned using a custom dataset of financial reports.
taste-visualizer
Generates animations to represent and visualize a user's taste across multi-entities of movies, books, TV shows using text-to-video Diffusion models.
tf_object_detection_multi_channels
Tutorial on How to change tensorflow object detection API to allow any number of input channels