How do I set up RoboSat with Docker for GPU training?

Use the command `docker run --runtime=nvidia -it --rm -v $PWD:/data --ipc=host mapbox/robosat:latest-gpu train` from the README, ensuring nvidia-docker is installed on the host system for GPU access.

Can RoboSat work with custom drone imagery instead of Mapbox satellite data?

Yes, you can bring your own imagery by converting it to Slippy Map tile format, as outlined in the 'Bring your own imagery' section, but alignment with masks is crucial.

What GPU is recommended for training models in RoboSat?

The README mentions using AWS p2/p3 instances or GTX 1080 TI GPUs, so mid-to-high-end GPUs with CUDA support are necessary for efficient training on large datasets.

RoboSat vs TensorFlow for satellite image segmentation?

RoboSat is a specialized pipeline with geospatial tools like OSM integration and post-processing, while TensorFlow offers general ML flexibility but requires more custom code for geospatial workflows.

How to export a trained model to ONNX in RoboSat?

Use the `rs export` tool to convert .pth checkpoints to ONNX format, enabling deployment in environments like Caffe2 or TensorFlow without Python dependencies.

Is RoboSat still supported or should I look for alternatives?

No, it's explicitly unmaintained per the README note, so consider active alternatives like Raster Vision or custom solutions if you need ongoing updates and support.

Open-Awesome

robosat

MITPythonv1.2.0

An end-to-end Python pipeline for semantic segmentation of aerial and satellite imagery to extract features like buildings and roads.

GitHub

2.1k stars385 forks0 contributors

What is robosat?

RoboSat is an open-source pipeline for semantic segmentation on aerial and satellite imagery. It extracts features such as buildings, parking lots, roads, and water by training neural networks on geo-referenced tile data. The project provides end-to-end tools for data preparation, model training, and post-processing to generate vector geometries.

Target Audience

Geospatial developers, data scientists, and mapping professionals working on automated feature extraction from satellite or aerial imagery. It is particularly useful for OpenStreetMap contributors and organizations needing scalable mapping pipelines.

Value Proposition

RoboSat offers a complete, extensible pipeline that abstracts away geospatial complexities with Slippy Map tiles. It integrates seamlessly with OpenStreetMap and supports custom data sources, making it a flexible alternative to proprietary geospatial AI services.

Overview

Semantic segmentation on aerial and satellite imagery. Extracts features such as: buildings, parking lots, roads, water, clouds

Use Cases

Best For

Automating building footprint extraction for OpenStreetMap
Training custom segmentation models on satellite imagery
Generating GeoJSON features from aerial drone surveys
Batch processing large-scale geospatial datasets
Creating training datasets for remote sensing applications
Deploying segmentation models in resource-constrained environments via ONNX

Not Ideal For

Projects requiring active maintenance and official support from the original developers
Real-time or streaming geospatial analysis applications
Teams without expertise in Docker, geospatial data, or machine learning

Pros & Cons

Pros

End-to-End Pipeline

Provides a complete toolkit from data preparation (e.g., rs extract for OSM data) to post-processing (e.g., rs features for GeoJSON output), streamlining feature extraction workflows.

Extensible Architecture

Supports custom imagery and masks through Slippy Map tile abstraction, allowing integration with diverse data sources as described in the extending section.

OpenStreetMap Integration

Seamlessly works with OSM for mask generation and deduplication, making it ideal for community mapping projects and automated updates.

Model Portability

Includes rs export for ONNX format, enabling deployment in resource-constrained environments like AWS Lambda without PyTorch dependencies.

Cons

Unmaintained and Abandoned

Explicitly marked as no longer maintained by Mapbox, with no bug fixes, updates, or security patches, posing risks for long-term projects.

Complex Docker Setup

Requires GPU-accelerated Docker containers with specific flags like --ipc=host and --runtime=nvidia, which can be error-prone for novice users.

Batch-Oriented Processing

Tools like rs predict and rs serve are designed for offline batch jobs, not real-time inference, limiting use in dynamic applications.

Frequently Asked Questions

Related Projects

detectron2

Detectron2 is a platform for object detection, segmentation and other visual recognition tasks.

Stars34,703

Forks7,937

Last commit20 days ago

EasyOCR

Ready-to-use OCR with 80+ supported languages and all popular writing scripts including Latin, Chinese, Arabic, Devanagari, Cyrillic and etc.

Stars29,980

Forks3,607

Last commit9 months ago

imgaug

Image augmentation for machine learning experiments.

Stars14,739

Forks2,451

Last commit2 years ago

meshroom

Node-based Visual Programming Toolbox

Stars12,948

Forks1,227

Last commit1 day ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub