Projects - Pranay

Featured Project

COMBAT: Conditional World Models

Conditional World Model for Behavioral Agent Training. A novel approach to training agents in complex environments. Presented at International Conference on Computer Vision 2025, Workshop on Reliable and Interactive World Models.

PyTorch VLMs Transformers ViT

Demo Code

Current Research Projects

Ongoing investigations pushing the boundaries of knowledge in Computer Vision and Machine Learning

Ongoing

Visual Place Recognition (VPR)

Developing novel deep learning models for cross-view image matching and image geolocalization. Investigating robust representations for challenging environments.

PyTorch Computer Vision Deep Learning Geolocalization

Completed Projects

Successfully concluded research initiatives with published outcomes

Completed

Cost Aggregation for Segmentation

Developed a cost aggregation framework for efficient segmentation using the Siglip2 Encoder. Focuses on efficient open vocabulary segmentation.

PyTorch VLMs Transformers ViT

Code

Completed

Occupancy Estimation (Thermal)

Privacy-preserving Occupancy Sensing Model trained on thermal images for real-time environments. Includes data collection and preprocessing pipelines.

PyTorch CV Raspberry Pi YOLOv8

Code Demo

Completed

Fine-Tuning LLaMA 2 (QLoRA)

Fine-tuned LLaMA 2-7B on Puffin Dataset using QLoRA and 4-bit quantization for improved efficiency and reduced memory consumption.

Colab Transformers PEFT HuggingFace

Code Demo

Completed

Transformer from Scratch

Implemented a Transformer model from scratch, focusing on attention mechanisms and efficient training techniques.

Transformer Colab PyTorch

Code