NumPy and PyMuPDF

Transform PDF documents into high-performance NumPy arrays for real-time computer vision and machine learning pipelines.

PyMuPDF (fitz) bridges the gap between static documents and dynamic data by converting PDF pages into NumPy ndarrays without intermediate file storage. Using the frombuffer method: operators can ingest a 300 DPI page into a (height, width, 3) RGB array in under 50 milliseconds. This integration is essential for feeding document pixels directly into OpenCV for image processing or PyTorch for layout analysis. It replaces slow disk-based workflows with a pure memory-to-memory transfer that scales across massive datasets.

https://pymupdf.readthedocs.io/en/latest/numpy.html