Introduction to Raster Vision
Raster Vision is a cutting-edge Python library and framework that facilitates the development of computer vision models tailored for satellite, aerial, and expansive imagery data, such as those captured by drones. The project offers specialized capabilities for chip classification, object detection, and semantic segmentation, utilizing PyTorch as its backend.
Functionality and Features
As a Library
Raster Vision extends a complete toolset for managing all components of a geospatial deep learning workflow. This includes the crucial steps of reading geo-referenced data, training models, making predictions, and recording the predictions in geo-referenced formats.
As a Low-code Framework
The project empowers users to effortlessly configure and execute machine learning experiments without needing in-depth expertise in deep learning. It enables users to manage tasks such as analyzing training data, creating training chips, training models, generating predictions, evaluating model performance, and packaging model files and configurations for seamless deployment.
Cloud Support
Raster Vision is compatible with cloud-based operations, supporting technologies like AWS Batch and AWS Sagemaker. This feature enhances scalability and accessibility for users who wish to run their experiments in cloud environments.
Getting Started
Installation Options
- Via pip: Easily install Raster Vision using Python's package manager.
pip install rastervision - Pre-built Docker Image: Utilize pre-built Docker images published on Quay.io. These images are updated with each new merge to the master branch.
- Build from Source: For users who prefer custom setups, clone the repository, then build and run the Docker image with provided scripts.
Usage and Tutorials
Raster Vision can be employed by non-developers as a low-code framework, simplifying complex tasks to configurable parameters. Beginners can start with the Quickstart guide, whereas more complex use cases are covered in the Examples section. Developers looking to integrate Raster Vision within their own software projects should explore the Usage Overview and Basic Concepts tutorials for deeper insight.
Community and Contributions
Raster Vision fosters an active community where users can engage and contribute:
- Discussion Forum and Mailing List: Join the Raster Vision community to discuss projects and share ideas.
- Contributing Guidelines: Contributions are welcome! Interested developers are encouraged to engage with the maintainers for significant updates or design changes, streamlining the contribution process. Contributors must sign a Contributor License Agreement.
License Information
Raster Vision operates under the Apache 2 license, ensuring an open and flexible use of the software. A comprehensive list of third-party licenses related to project dependencies is also available.
By equipping users with the tools, simplicity, and support they need, Raster Vision stands as an essential resource for developing advanced geospatial computer vision applications.