My Projects
Goal Conditioned Framework
A goal-conditioned reinforcement learning framework with DDPG, SAC, TD3 and TQC for the Panda-Gym environment. Trains robotic arm for reaching, pushing, sliding and pick-place tasks.
AMSTRA
Autonomous drone navigation framework with real-time object detection and spatial tagging. Uses YoloV8 and SORT for tracking with 3D spatial mapping capabilities.
GenIText
Flexible framework for generating high-quality Image-Text pairs for fine-tuning generative models like Stable Diffusion and DALL-E using open-source VLLMs.
MujRL
Implementation of TD3 and SAC reinforcement learning algorithms for MuJoCo environments including Ant, HalfCheetah, Hopper, and Walker for robotic control tasks.
Efficient Deep Q-Learning
Resource-conscious DQN framework for Atari games optimized for consumer hardware. Achieves 57% better performance than standard DQN with efficient scaling.
CNN PoseNet Suite
6-DoF camera pose estimation using deep learning models on 7-Scenes dataset. Compares different CNN backbones for performance tradeoffs in pose estimation.
UNet Image Segmentation
Semantic segmentation of aerial images using UNet to classify land, water, buildings, roads, and vegetation in Dubai imagery.
UNet Image Denoising
Educational repository for U-Net image denoising with support for consistent and varying noise levels. Performs general image denoising with room for detail recovery improvements.
Neural Radiance Field
Educational implementation of NeRF for novel view synthesis. Renders objects at unseen angles with mask losses for improved foreground/background rendering.
ESRGAN
Enhanced Super-Resolution Generative Adversarial Network for high-fidelity image super-resolution. Trained on HD images for generalization across various contexts.
Cognitus Chess Engine
Python-based chess engine with Alpha-beta pruning. Estimated ELO rating of 1550-1750, performing at 4-7 depth evaluations due to computational constraints.
Canny Edge Detection
Complete implementation of Canny Edge Detection algorithm with enhancements for accuracy and speed. Includes FFT-based convolution and configurable parameters.
Traffic Control RL
Deep reinforcement learning for traffic signal optimization using SUMO simulations. Modified DuelDQN with weight transfer for multi-intersection scenarios.
Lunar Lander RL
Reinforcement learning on LunarLander-v2 using DQN and Actor-Critic methods. Actor-Critic converges 4x faster than DQN in continuous action spaces.
Stable Diffusion Fine-tuning
Educational framework for fine-tuning Stable Diffusion models via HuggingFace. Supports conditional and unconditional training with lightweight implementation.
FFT Audio Visualization
Fast Fourier Transform for audio signal decomposition and visualization. Converts MP3 audio to frequency-based visualizations compiled into matching MP4 videos.
Image Clustering with Resnet18
Image clustering with Resnet18. The project focuses on clustering the images into clusters based on their features using KMeans clustering. The project supports html display of 3D visualiazation results