Gnana Prakash's repositories
Statistical-Methods-In-AI-M23
Coursework done as part of the Statistical Methods in AI course offered in Monsoon 2023 by Prof. Ravi Kiran Sarvadevabhatla, IIITH. Topics covered include KNNs, Decision Trees, Dimensionality Reduction, Gaussian Mixture Models, Bagging, Boosting, MLP Classifiers and Regressors, Logistic Regression, Kernel Density Estimation and Hidden Markov Models
awesome-Visual-Place-Recognition
A curated list of awesome Visual Place Recognition papers
Camera_LiDAR_Calibration
Implemented an end-to-end calibration pipeline starting with an uncalibrated camera. Utilized checkerboards along with Zhang's camera calibration method to calibrate the camera. Applying the PnP algorithm between manually annotated correspondences between LiDAR maps of the checkerboard and the corresponding images gives the extrinsic parameters
gpsh
A custom shell implemented in C that integrates essential functionalities such as echo, cd, ls, pwd, history, pinfo, and signal handlers (CTRL-C, CTRL-Z). Additional features include tab-based autocomplete, support for foreground and background processes, inter-process communication using piping, and custom functionalities like "discover"
Greddiit
A fullstack application inspired by Reddit utilizing ReactJS, Material-UI (MUI), Node.js, Express, and MongoDB, with Docker for deployment. Key features include user authentication using JSON Web Tokens (JWT), password reset functionality, user and moderator roles, creation of subgreddits, posts, comments, votes, followers, and email notifications
IPL_Miniworld_DBMS
A CLI-based database application built using MySQL and Python for managing and displaying IPL data, granting users access to information about players, matches, runs, and wickets. Focused on different entity types in the database design, achieving a final relational model in 3NF
MNIST-Classification
Implementation of ViT (Vision Transformer) and LeNet from scratch using PyTorch for the MNIST classification task. Ran multiple experiments to understand the impact of dataset size, network depth, optimizer, learning rate, and batch size on model performance
Mobile-Robotics-M23
Coursework done as part of the Mobile Robotics course offered in Monsoon 2023 by Prof. Madhava Krishna, IIITH
Object_Detection_And_Tracking
Explored Classical and DL based approaches - Employed the Viola-Jones facial classifier for face detection and tracking, adjusting hyperparameters to optimize the quality of detections and tracks. Utilized YOLOv8 models for duck detection, running experiments to analyze the impact of model size, augmentations, dataset size and weight initialisation
Pose_Shape_Optimisation_Pipeline
Pipeline to accurately and efficiently compute the shape and orientation (pose) of a car given its RGB image by using keypoints generated from an hourglass network after optimising the estimates using the Ceres solver
UNet_Segmentation
Segmentation pipeline that uses a U-Net backbone to perform segmentation on the Cityscapes dataset. Conducted experiments to analyse the impact of the skip connections of the U-Net on the quality of the segmentation masks. These masks are also qualitatively analysed using the Intersection-over-Union (IoU) metric
xxv6
Extension of the xv6 operating system designed by MIT. Includes features such as strace (to keep track of invoked syscalls), sigalarm and sigreturn (to periodically alert processes using up CPU time), four types of schedulers - FCFS, Lottery-based, Priority-based and Multi-Level Feedback Queue schedulers and Copy-On-Write fork