n_stories

Story-count detector using the BRAILS++ EfficientDet-D4 model.

Ports the BRAILS++ floor-detector runtime into openplaces, replacing cv2 with PIL and removing the brails package import. All post-processing (threshold tuning, nested-box removal, stack grouping, middle-region selection) is reproduced verbatim from the BRAILS++ source.

The EfficientDet inference engine is self-contained in efficientdet_lib.infer.Infer. Model weights are downloaded from Zenodo on first use and cached under cfg.models_dir / ‘external’ / ‘brails’.

Classes

NStoriesDetector

Detect the number of stories in buildings from street-view images.

Module Contents

class openplaces.io.enricher.detectors.n_stories.NStoriesDetector(model_path: str | pathlib.Path | None = None)

Detect the number of stories in buildings from street-view images.

Uses the BRAILS++ EfficientDet-D4 model (downloaded from Zenodo on first use) plus the original BRAILS++ post-processing pipeline. The inference engine is bundled with openplaces.

Parameters:

model_path – Path to the pre-trained .pth weights file. When None, the pretrained model is cached under cfg.models_dir / ‘external’ / ‘brails’. Existing files in the two previous external-data locations are still used.

predict(images: openplaces.io.scrapers.types.ImageSet, checkpoint: openplaces.io.enricher.detectors.checkpoint.PredictionCheckpoint | None = None) → dict

Predict the number of stories for each image in images.

Parameters:

• images – Collection of street-view images to analyse.

• checkpoint – When given, previously checkpointed predictions are reused, new ones are persisted periodically, and an interrupted run resumes where it left off.

Returns:

Mapping from the same keys as images.images to an integer story count, or None when the image file does not exist.

Return type:

dict