Cropper

matmmextract.inference.cropper

Crop detected panels from images using the JSON files produced by run().

Output naming convention

<stem>_<label>.jpg — highest-confidence detection for that label. When multiple detections share a label, only the one with the highest score is saved. These names are matched by dataset_builder’s IMG_RE pattern.

class CropResult(n_crops: 'int' = 0, n_skipped: 'int' = 0, n_no_detections: 'int' = 0, output_dir: 'str' = '', failed: 'list[str]' = <factory>)

failed: list[str]

n_crops: int = 0

n_no_detections: int = 0

n_skipped: int = 0

output_dir: str = ''

crop(image_dir: str | Path, json_dir: str | Path, output_dir: str | Path, limit: int | None = None, verbose: bool = True) → CropResult

Crop all detected panels and save them as individual JPEG files.

Parameters:

• image_dir – Directory containing the original flat images.

• json_dir – Directory containing per-image .json files from run(). _summary.json is automatically skipped.

• output_dir – Directory where cropped panel images are saved.

• limit – Process at most this many JSON files (useful for testing).

• verbose – Print progress.