An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
import threading
import sopdf
import numpy as np
import cv2
# PDFium requires process-wide exclusion, even for separate documents.
_PDFIUM_LOCK = threading.Lock()
def pdf_to_images(pdf_bytes: bytes, dpi: int = 150) -> list[np.ndarray]:
"""Convert PDF bytes to a list of OpenCV images (numpy arrays, BGR format)."""
with _PDFIUM_LOCK:
images = []
with sopdf.open(stream=pdf_bytes) as doc:
for page in doc.pages:
png_bytes = page.render(dpi=dpi)
nparr = np.frombuffer(png_bytes, np.uint8)
cv_image = cv2.imdecode(nparr, cv2.IMREAD_COLOR)
images.append(cv_image)
return images