API Reference

Everything below is importable from the top-level rapidtools package as well as from the subpackage shown. Heavy dependencies load on first use.

Core objects

PhysicalAsset	Represents a tangible, real-world entity.
PhysicalAssetCollection	A specialized container for collection of PhysicalAsset objects.
ImageAsset	A single image asset, its location, metadata, and segmentation info.
ImageCollection	A container class for managing a group of ImageAsset objects.
BoundingBox	Represents an axis-aligned rectangular region.
PolygonRegion	Represents a geographical region defined by a single polygon shell.
Observation	One sighting of an object in a street-level image.
OperationCancelled	Raised by a component when its cancel_event has been set.
is_cancelled	Return True if cancel_event exists and has been set.
raise_if_cancelled	Raise OperationCancelled if cancel_event has been set.

Pipeline and analysis

Pipeline	A smart sequence of processing steps applied to a PhysicalAssetCollection.
Stage	Execution order of pipeline steps (lower runs first).
PipelineStep	Structural type of a pipeline step.
AssetAnalyzer	Describe every asset in a collection with a vision-language model.
RateLimitPolicy	Shared back-off applied by AssetAnalyzer to hosted APIs.
DetectionVerifier	Confirm that each detected object is what its asset type claims.
DuplicateResolver	Merge pairs of nearby objects that a model says are one vehicle.

Imagery extractors

AerialImageryExtractor	Component that extracts aerial imagery patches for a collection of assets.
BingOrthomosaicExtractor	Component that extracts and synthesizes aerial imagery from Bing Maps Tiles.
GoogleOrthomosaicExtractor	Component that stitches Google Maps satellite tiles into a GeoTIFF.
GoogleStreetViewImageExtractor	Street-view extractor that fetches keyless Google Street View crops of assets.
MapillaryImageExtractor	Street-view extractor that fetches unobstructed Mapillary panoramas of assets.
MapillaryObjectImageExtractor	Crop each detected object out of the street-level images that saw it.
MapillaryLabelMapper	Map natural-language class names to exact Mapillary segmentation labels.

Detection, segmentation and regularization

SAM3OrthoFeatureExtractor	Pipeline component that uses local SAM 3 to discover real-world assets.
MapillaryFeatureExtractor	Discover objects along a Mapillary survey and place them on the ground.
SAM3ImageSegmenter	Pipeline component that uses local SAM 3 to segment images attached to assets.
BuildingRegularizer	Pipeline component to regularize and refine preliminary building polygons.
RoadwayRegularizer	Pipeline component to regularize jagged raster-to-vector road polygons.
vectorize.mask_to_wgs84_polygons	Vectorize a binary mask into valid WGS84 polygons.

Street-level localisation

How sightings in street-level images become positioned objects: the camera model, tracking and triangulation, covariance-aware merging, the ray-voting baseline and appearance re-identification.

street_localization.localize	Fill in the single-view ground position of an observation.
street_localization.observation_angles	Bearing to an observation and elevation of its ground contact.
street_localization.pixel_bearing	Unit direction in the camera frame of a normalised image point.
street_localization.world_ray	Rotate a camera-frame direction into the world (east, north, up) frame.
street_localization.rotation_matrix	Rotation matrix of an axis-angle vector (Rodrigues' formula).
street_localization.intersect_bearings	Least-squares intersection of bearings from several camera positions.
street_localization.estimate_ego_mask	Find detections of the collection vehicle itself.
street_localization.thin_frames	Keep one frame per spacing_m metres along each sequence.
street_tracking.discover_objects	Turn the sightings of one class into object estimates.
street_tracking.suppress_duplicate_sightings	Drop a detector's duplicate outlines within each frame.
street_tracking.prune_unwitnessed	Drop estimates that nearby frames should have seen but did not.
street_tracking.CameraFrame	Where one survey frame was taken and what it could see.
street_tracking.track_sequence	Link the sightings of one sequence into tracks, one per object.
street_tracking.merge_by_rays	Merge estimates whose lines of sight point at the same object.
street_tracking.triangulate_track	Position a track from its rays, or None when it is not a static object.
street_tracking.merge_estimates	Fuse estimates of the same object; keep different objects apart.
street_tracking.merge_pieces	Fold objects that are really pieces of a neighbour into that neighbour.
street_tracking.sighting_shows	Whether a sighting can be of an object at target.
street_tracking.vote_rays	Locate objects as peaks of accumulated ray votes on a ground grid.
street_tracking.ObjectEstimate	One candidate object: a position with uncertainty and its sightings.
reid.AppearanceEmbedder	L2-normalised image embeddings from a Hugging Face vision backbone.
reid.merge_by_appearance	Group candidates that are close together and look alike.
reid.pairs_to_compare	List the candidate pairs whose appearance is worth comparing.
reid.views_for_reid	The sightings of a candidate worth cropping for re-identification.
reid.polygon_width_px	Width of a detection outline in pixels of the image it was drawn on.
reid.crop_observation	Cut the axis-aligned crop of a detection polygon out of its source image.
reid.recommended_min_similarity	Cosine-similarity threshold that suits a backbone.

Provider-specific analyzers

These predate AssetAnalyzer and remain for compatibility. New code should use AssetAnalyzer(load(provider, ...)).

GeminiAssetAnalyzer	Deprecated: use AssetAnalyzer(load('gemini', ...)).
ClaudeAssetAnalyzer	Deprecated: use AssetAnalyzer(load('claude', ...)).
OpenAIAssetAnalyzer	Deprecated: use AssetAnalyzer(load('openai', ...)).
MuseSparkAssetAnalyzer	Deprecated: use AssetAnalyzer(load('muse_spark', ...)).
QwenAssetAnalyzer	Deprecated: use AssetAnalyzer(load('qwen', ...)).
Gemma4AssetAnalyzer	Deprecated: use AssetAnalyzer(load('gemma4', ...)).
LlamaVisionAssetAnalyzer	Deprecated: use AssetAnalyzer(load('llama', ...)).
MuseGlimmerAssetAnalyzer	Deprecated: use AssetAnalyzer(load('muse_glimmer', ...)).
QwenVisionAssetAnalyzer	Deprecated: use AssetAnalyzer(load('qwen_vl', ...)).
HFVisionAssetAnalyzer	Deprecated: use AssetAnalyzer(load('hf', ...)).

Models

The provider registry rapidtools.models.PROVIDERS maps each registry key to a ModelInfo record and is always importable without PyTorch.

load	Instantiate a model wrapper from its registry key.
list_providers	Return registry keys, optionally filtered by kind.
catalog	Return the curated model catalogue of a provider (no network).
get_model_class	Import and return the wrapper class registered under provider.
ModelInfo	Static description of a model backend.
GenerationConfig	Text-generation options shared by every vision-language wrapper.
SegmentationConfig	Options for promptable segmentation models such as SAM 3.
ModelOutput	A universal container for multimodal AI model outputs.
BaseInferenceModel	The top-level contract for all AI models (API or local).

Hosted models

GeminiInference	Google Gemini generateContent client for image + text prompts.
ClaudeInference	Anthropic Claude Messages API client for image + text prompts.
OpenAIInference	OpenAI Chat Completions client for image + text prompts.
MuseSparkInference	Meta Model API client for the Muse Spark family.
QwenInference	Alibaba Cloud Model Studio (DashScope) client for hosted Qwen models.

Local models

Gemma4Inference	Local inference model for Google's Gemma 4 vision-language models.
LlamaVisionInference	Local inference with Meta's Llama 4 and Llama 3.2 Vision models.
MuseGlimmerInference	Local inference with Meta's open-weight Muse Glimmer vision-language model.
QwenVisionInference	Local inference with open-weight Qwen vision-language checkpoints.
HFVisionInference	Run any Hugging Face vision-language checkpoint that ships a chat template.
SAM3Inference	Text-prompted instance segmentation with SAM 3.

Data sources

OrthomosaicReader	Client for extracting high-resolution imagery from local orthomosaic rasters.
orthomosaic_reader.PatchGeoref	Native georeferencing of an image patch read from a raster.
GoogleAerialImageExtractor	Client for extracting satellite imagery from Google Maps tiles.
BingAerialImageExtractor	Client for extracting high-resolution imagery from Bing Maps tiles.
GoogleStreetViewClient	Keyless client for locating, downloading and cropping Street View panoramas.
StreetViewPanorama	Metadata describing one Google Street View panorama.
MapillaryClient	Fetch and store Mapillary images by ID or geographic bounding box.
MapillaryLabels	Namespace of official Mapillary segmentation label strings.
TileUtils	Utility methods for Web Mercator/Mapbox Vector Tile operations.

Datasets, configuration and authentication

datasets.download_dataset	Download all files associated with one or more datasets from the registry.
config.configure_logging	Attach a console handler to the rapidtools logger.
config.get_configured_session	Create a requests Session with retry logic, timeouts, and headers.
config.resolve_alias	Pop a deprecated keyword alias from kwargs and return the value.
models.api_base.api_session	Return a session with the retry policy used for provider API calls.
auth.login	Authenticate this machine with the Hugging Face Hub.
auth.ensure_huggingface_login	Log in only when no Hugging Face token is cached yet.

Graphical interface

launch_asset_analysis_app	Start the GUI server and open it in the default browser (blocks).
AssetAnalysisWorkflow	Orchestrates asset detection and VLM inference on an aerial raster.
DetectionSettings	User-configurable options for the asset detection stage.
StreetDetectionSettings	Options for AssetAnalysisWorkflow.discover_street().
RegionImagerySettings	Options for AssetAnalysisWorkflow.download_basemap().
InferenceSettings	User-configurable options for the VLM inference stage.
AssistSettings	Options for AssetAnalysisWorkflow.assist().
NotificationConfig	Where notifications are sent from.
Notifier	Send job summaries by email or webhook.
JobSummary	What a finished job looks like to the recipient.
PromptSpec	Everything needed to assemble a prompt in the rapidtools sample layout.
OutputField	One line of the REQUIRED OUTPUT FORMAT block.
RubricEntry	Description of one class of the primary categorical field.
assemble_prompt	Render a PromptSpec in the layout of the rapidtools sample prompts.
PromptAssistant	Use a rapidtools model backend to help write a PromptSpec.