RasterFlow run AI on satellite imagery at any scale RasterFlow is the production-grade inference engine for Earth intelligence. Bring your own PyTorch model or use the built-in library. Point it at any imagery source. RasterFlow returns structured spatial features, ready for spatial joins in WherobotsDB. The bridge between what satellites capture and what your AI reasons about. START BUILDING Run inference on satellite imagery at planetary scale No custom computer vision pipeline required. Combine raster and vector in one pipeline Satellite-derived features flow into spatial joins in WherobotsDB. From built-in models to bring-your-own Fields of the World, Tree Canopy Height, SAM3, or any PyTorch model. Unlock Geospatial Intelligence with AI From satellite imagery analysis to production spatial features, in one pipeline. AI for remote sensing imagery Hosted Models Raster and vector analysis Automate and scale Apply AI models to remote sensing imagery Use state of the art AI models on remote sensing imagery for object detection, segmentation and classification. Or bring your own model. Learn More Leverage out of the box models on your imagery See all the models that Wherobots supports on Hugging Face. Learn More Combine raster and vector analysis Seamlessly store model inference results in Wherobots for further analysis and enrichment with vector or tabular data. All within a single platform and single development experience through Python or Spatial SQL. Learn More Automate and scale Scale your analysis from small-scale prototyping to planetary-scale production workflows, only pay for what you use. Learn More Benefits of AI for Earth Built-in models, or bring your own Fields of the World, Tree Canopy Height, Road Segmentation, SAM3. Or bring any PyTorch model. Learn more Solution notebooks for real-world use cases Agriculture, land cover, and infrastructure detection. Start from working code. Learn More From a single tile to continental coverage Automatic mosaicking, cloud removal, and distributed processing. Scales with your data. Learn More Integrated with WherobotsDB Inference outputs flow into WherobotsDB as vector features for spatial joins and downstream analytics. Learn More GPU-optimized serverless inference Cost-effective at any scale. No infrastructure to manage. Learn More Instant results in open formats Zarr, GeoParquet and Apache Iceberg output. Query outputs immediately in WherobotsDB or your lakehouse. Learn More Turn satellite imagery into structured spatial features RasterFlow detects buildings, roads, vegetation, water, and land cover from satellite imagery. No manual classification. No custom computer vision pipelines. Point RasterFlow at imagery, get analysis-ready vector features back. START BUILDING Satellite imagery analysis in action Object detection, field boundary extraction, land cover classification, and road segmentation, powered by RasterFlow. Airplanes detected with SAM3 Find airplanes from open satellite imagery using Wherobots and the SAM3 model with a single text prompt. Solar farms detected with RasterFlow Detect solar farms in open satellite imagery using Wherobots and the Allen Institute Satlas models. Getting started with RasterFlow Run GeoAI models with a few lines of Python. RasterFlow handles distributed inference, mosaicking, and vectorization. Post-process or join results in WherobotsDB. Python Mosaicking Inference Learn More model_output_index = rf_client.build_and_predict_mosaic_recipe( # Path to an AOI in GeoParquet or GeoJSON format aoi = aoi_path, # Date range for Sentinel-2 imagery start = datetime(2023, 1, 1), end = datetime(2024, 1, 1), # Coordinate Reference System for the output mosaic crs_epsg = 3857, # Model recipe for inference (FTW in this case) model_recipe = ModelRecipes.FTW ) Bring your own model Use your own PyTorch model with RasterFlow. Load your PyTorch PT2 model archive from Hugging Face or S3. Run distributed segmentation, detection, or regression against mosaics from satellite or aerial imagery at planetary scale. PyTorch Custom ML Models Hugging Face Learn More custom_inference_config = InferenceConfig( model_path = hugging_face_model_path, actor = MosaicToMosaicActorEnum.REGRESSION_PYTORCH, features = ["red", "green", "blue"], labels = ["canopy_height"], ... ) Build cloud free mosaics Build cloud free mosaics from Sentinel-2 in Zarr format. Choose from different compositing options including median, best scene or pre-configured settings for planting season and harvest season. Mosaicking Sentinel-2 Zarr Learn More mosaic_index = rf_client.build_mosaic( # Dataset for harvest season datasets=[DatasetEnum.S2_MED_HARVEST], # Path to your AOI aoi=aoi_path, # Date range for Sentnel-2 imagery start=datetime(2024, 1, 1), end=datetime(2025, 1, 1), # Coordinate Reference System for the output mosaic crs_epsg=3857 ) FAQ What is RasterFlow? RasterFlow is Wherobots’ serverless inference engine for satellite imagery analysis and Earth intelligence. It builds mosaics from multiple raster data sources, runs inference with PyTorch computer vision models, and vectorizes results into spatial features. RasterFlow supports semantic segmentation, instance segmentation, object detection and regression at planetary scale. The engine manages its own compute resources and autoscales based on area, time range, dataset density, and model complexity. What models does RasterFlow include? RasterFlow includes curated, open-source models for common Earth observation tasks. Built-in models include Fields of the World for agricultural field boundary detection from Sentinel-2 imagery, Meta CHM v1 for canopy height estimation, Tile2Net for urban infrastructure detection (sidewalks, crosswalks, pedestrian pathways), ChesapeakeRSC for rural road detection and Meta’s Segment Anything 3 (SAM3) for instance segmentation and object detection from text prompts. RasterFlow also supports bring-your-own-model: register your PyTorch PT2 model, upload it to your cloud storage, and run distributed inference through the RasterFlow API or as an automated Job Run. What imagery does RasterFlow work with? RasterFlow includes built-in satellite and aerial imagery datasets. Sentinel-2 provides worldwide coverage at 10m to 30m resolution with 5-day revisit rates. RasterFlow automatically applies cloud and quality filtering when building Sentinel-2 mosaics, masking out cloud pixels, shadows, and sensor anomalies. NAIP (National Agriculture Imagery Program) provides high-resolution imagery (30cm to 1m) across the continental United States. RasterFlow can also build mosaics and run inference on your own GeoTIFF imagery stored in cloud storage. Get started RasterFlow is part of Wherobots, the AI Context Engine for the Physical World. TALK TO US START BUILDING
