Geospatial Image Segmentation with Meta's Segment Anything Model 3

Seg2Map is an interactive web map app for geospatial image segmentation using deep learning

A repository for segmenting satellite images using VM-UNet, a variant of the Mamba architecture originally designed for medical image segmentation. This project adapts VM-UNet for geospatial applications, optimizing it for multi-channel satellite imagery.

Identifying point-of-interests within satellite images

A satellite image segmentation project using DeepLabV3+ for land cover classification. The project integrates Google Earth Engine (GEE) for data collection, applies advanced image augmentation, and uses TensorFlow to train a semantic segmentation model. Ideal for geospatial analysis and remote sensing applications.

LAM (Locate Anything Model) — standalone prompt based image segmentation + geospatial tiling as a service.

A deep-learning project for performing semantic segmentation on geospatial imagery, using architectures such as U-Net, LinkNet and DeepLab v3+ to extract meaningful features from overhead or satellite image datasets.

Command line interface for geospatial image segmentation

SAR-DeepSeg is an end-to-end pipeline for semantic segmentation of Synthetic Aperture Radar (SAR) images. It employs deep learning models to process and segment SAR imagery, facilitating accurate interpretation for applications such as land cover segmentation, environmental monitoring, and disaster response, enhancing geospatial analysis.

Island Detection using Transformer Neural Networks – A deep learning project utilizing Vision Transformers (ViTs) for island segmentation from satellite images. Useful for geospatial analysis, environmental monitoring, and automated mapping. Built with PyTorch/TensorFlow and satellite imagery datasets.

Flask web application integrated with a YOLOv11-seg deep learning model to perform real-time instance segmentation for waste classification, Engineered an end-to-end image processing pipeline that provides users with immediate category predictions for uploaded media and Leveraged the Google Maps API to embed geospatial features

No description available

Segmenting satellite imagery and geospatial data.

"Classical" object-based image segmentation (OBIA) for geospatial data

Deep learning model for precise satellite image segmentation and geospatial analysis

This project is a Flutter-based mobile application designed for land resource management, featuring geospatial visualization, image capture, and AI-powered image segmentation.

MVP of deep learning software for satellite image classification and geospatial anomaly visualization, using a ResNet-18 backbone with custom head for multi-class land cover segmentation.

Implemented a U-Net model in PyTorch for aerial image segmentation to detect road networks from satellite data. Designed custom data pipelines and applied augmentations to improve segmentation performance, demonstrating practical skills in computer vision for geospatial applications.

DeepCoral is a full-stack AI web system that uses deep learning–based image segmentation to automatically estimate coral lifeform cover from quadrat images, with geospatial tagging and data visualization for reef monitoring and analysis.

A deep learning–based satellite image change detection system using U-Net and multi-temporal Sentinel-2/Landsat imagery to detect deforestation, urban growth, and flood-affected regions through semantic segmentation and geospatial analysis.

End-to-end Python/YOLOv8 pipeline for solar rooftop detection from satellite imagery. Uses segmentation, image tiling/merging, and geospatial (GIS) post-processing with Rasterio/Geopandas to generate georeferenced GeoJSON polygons with calculated area.

This project focuses on **image segmentation for water detection** using harmonized multispectral satellite imagery from **Landsat and Sentinel-2**. It tackles challenges across the pipeline—from data preprocessing to model deployment—providing a robust deep learning solution for geospatial applications.

AI-powered system that analyzes drone imagery to detect rural features like buildings and roads. It uses deep learning for segmentation, processes images in chunks, and visualizes results as GeoJSON on an interactive map, enabling efficient rural planning and real-time geospatial insights.

This project enables zero-shot satellite image segmentation using vision-language models CLIP, SAM). Simply describe what you want to find—forests, water, buildings—and the model segments it without training. Ideal for environmental monitoring,disaster assessment,agriculture, and urban analysis. Making advanced geospatial AI accessible to everyone.

A Flask web application integrated with a YOLOv11-seg deep learning model to perform real-time instance segmentation for waste classification , An image processing pipeline that provides users with immediate category predictions for uploaded media , the Google Maps API for geospatial features, enabling users to locate nearby recycling centers

Developed a web application that uses on-device GPU/NPU for interactive semantic segmentation on images loaded using WMS service. Challenge: To ensure the system is user-friendly and accessible, even for non-technical users. It should provide data export in geospatial format of user selected features. should make maximum utilisation of ondevice