Rohit Gandikota's Projects
arXiv LaTeX Cleaner: Easily clean the LaTeX code of your paper to submit to arXiv
Automatic Image Quality Analysis (AIQA) has become a very crucial module in remote sensing industry. With increasing competition and institutions that provide remote sensing images, the quality of images provided to the users has a huge impact.
This project shows the usage of hugging face framework to answer questions using a deep learning model for NLP called BERT. This work can be adopted and used in many application in NLP like smart assistant or chat-bot or smart information center.
If you are using an intranet network and have no means to communicate, use this python code to make a server where clients in the same network can chat. This is a pet project of me and Samvram Sahu.
This project harnesses deep reinforcement learning to detect cars in aerial images
This project uses deep RL to train an agent that can play Atari game named Space Invaders.
This repository contains our initial experiments to study code deprecation in codeLLMs
Importance of eigen directions in deep neural networks
Erasing Concepts from Diffusion Models
Satellite data feature matching flow
Generative Models to hide Audio inside Images using custom loss functions and Spectrogram Analysis
Variational Autoencoder-Generative Adversarial Network (VAE-GAN) to hide data inside images
This is a preliminary attempt on hiding video data inside images using deep learning. We design a custom adversarial network with custom losses and additional discriminator. We call this multi-discriminator and multi-objective training framework.
Recognition using Tensorflow
This code can be used for finding the angle that the image has been rotated by. Especially is tested on satellite data where geo-referencing rotates the image.
Understanding how DQN agents can play atari games like pingpong
This work discusses how high resolution satellite images are classified into various classes like cloud, vegetation, water and miscellaneous, using feed forward neural network. Open source python libraries like GDAL and keras were used in this work. This work is generic and can be used for satellite images of any resolution, but with MX band sensors.
Satellite images are being extracted by Space organisations around the world. One of the objectives of the agencies is to mask out sensitive area for security reasons. For this purpose, there are certain predefined shapefiles in terms of Latitude and Longitude of earth. These shapefiles are to be used on the raw images that are not Geo-referenced and hence are ought to be handled in the co-ordinate axis (Scan-Pixels) . For this reason, a file that maps certain scan-pixels to latlons is provided so that one can project all the scanpixels to latlons and mask the area using shapefiles. This is specifically not opted as the product/satellite image has to dessiminated to users in the scan-pix. For this reason we propose machine learning techniques to accomplish this challenging task.
Text-based and Voice-based search for satellite data ordering will massively improve user usability in terms of time spent and ease. This work focuses on satellite specific lingo and uses databases to search for data.
Open Source Computer Vision Library
We explore the concept of progressive growth of network layers in denoising diffusion probabilistic models.
The goal of this assignment is to learn and practice sets, dictionaries and objects. This assignment has two parts, first about dictionaries and sets and the second one about objects. Put all the required documents into a folder called a2_xxxxxx where you changed xxxxxx to your student number, zip that folder and submit it as explained in Lab 1. In particular, the folder should have the following three files: a5_part1_xxxxxx.py, a5_part2_xxxxxx.py and a5_part2_testing_xxxxxx.txt where you changed xxxxxx to your student number.
This project deals with the real time cloud detection of the ongoing acquisition data of satellite images. For this end, we use a simple and light MLP for classification of the image pixels. This work can classify the satellite images of size ranges till 64000 pixels width.
AI Research
Build a Jekyll blog in minutes, without touching the command line.
A Conditional Patch GAN for synthesis of optical images from SAR data as a 24X7, all weather disaster surveillance
This work
Generative Model to generate Map layers from Satellite Data
Concept Sliders for Precise Control of Diffusion Models
API for scrapping news on stock market for sentiment analysis and stock prediction