raster-vision: a geospatial computer vision framework for building ML models on satellite and aerial imagery

What it solves

Raster Vision provides a comprehensive framework for applying computer vision to large-scale geospatial imagery, such as satellite, aerial, and drone data. It simplifies the process of building, training, and deploying deep learning models for tasks like object detection and semantic segmentation on geo-referenced data.

How it works

It operates as both a library and a low-code framework. As a library, it provides utilities for reading geo-referenced data, training models, and writing predictions back into geospatial formats. As a framework, it allows users to configure a machine learning pipeline—including data analysis, chip creation, model training, and evaluation—without needing to be deep learning experts. It uses PyTorch as its backend and supports cloud execution via AWS Batch and AWS SageMaker.

Who it’s for

• Developers who want to integrate geospatial CV tools into their own code.
• Non-developers or geospatial analysts who need a low-code way to configure and run ML experiments on imagery.

Highlights

• Built-in support for chip classification, object detection, and semantic segmentation.
• Full geospatial workflow support, from reading geo-referenced data to outputting predictions in geo-referenced formats.
• Low-code configuration for repeatable ML pipelines.
• Native integration with AWS Batch and AWS SageMaker for cloud scaling.