Standard OCR fails on messy slide labels. OpenCV preprocessing plus PaddleOCR got us to 95% accuracy on tiny, low-contrast text.
Tagged Python, Computer Vision, OCR, OpenCV, Deep Learning, Image Processing.