Types API¶

Core type definitions and data structures.

BBox¶

Unified bounding box representation with automatic format conversion.

from pipeline.types import BBox

# Create from xyxy coordinates
bbox = BBox(x0=100, y0=50, x1=500, y1=200)

# Convert from different formats
bbox = BBox.from_yolo([0.5, 0.3, 0.4, 0.2], width=1000, height=800)
bbox = BBox.from_mineru([100, 50, 400, 150])  # xywh
bbox = BBox.from_pypdf_rect([100, 550, 500, 600], page_height=792)

# Export
xywh = bbox.to_xywh_list()  # [100, 50, 400, 150]

Block¶

Represents a detected layout block with metadata.

from pipeline.types import Block, BBox

block = Block(
    type="text",
    bbox=BBox(100, 50, 500, 200),
    detection_confidence=0.95,
    order=0,
    text="Extracted text",
    source="doclayout-yolo"
)

Protocols¶

Detector¶

class Detector(Protocol):
    def detect(self, image: np.ndarray) -> list[Block]:
        ...

Sorter¶

class Sorter(Protocol):
    def sort(self, blocks: list[Block], image: np.ndarray) -> list[Block]:
        ...

Recognizer¶

class Recognizer(Protocol):
    def process_blocks(self, image: np.ndarray, blocks: Sequence[Block]) -> list[Block]:
        ...

    def correct_text(self, text: str) -> str | dict[str, Any]:
        ...

Full API Reference

Detailed API reference coming soon. See BBox Formats for conversion examples.