Source code for torchocr.io
"""I/O helpers for reading OCR inputs."""
from pathlib import Path
from torch import Tensor
from torchvision.io import ImageReadMode, decode_image
from .pdf import load_pdf
[docs]
def load_image(path: str | Path, mode: ImageReadMode = ImageReadMode.RGB) -> Tensor:
"""Load an image as a ``(C, H, W)`` ``uint8`` tensor, RGB by default.
Grayscale scans and RGBA screenshots are converted to the 3-channel RGB
that pretrained models and :class:`~torchocr.OCRPipeline` expect; pass
``mode=ImageReadMode.UNCHANGED`` to keep the file's own channels.
"""
return decode_image(str(path), mode=mode)
__all__ = ["load_image", "load_pdf"]