### Expected behaviour Using `cv2.imread` from 4.5.5.62 on a PNG image with rotation (EXIF orientation) results in correctly loaded image: ``` >>> import cv2 >>> a = cv2.imread("ex.png") >>> a.shape (844, 1500, 3) ``` ### Actual behaviour Using `cv2.imread` from 4.5.5.64 on a PNG image with rotation results in image loaded with rotation ignored: ``` >>> import cv2 >>> a = cv2.imread("ex.png") >>> a.shape (1500, 844, 3) ``` Unfortunately I cannot provide image in question as it contains sensitive information. ### Steps to reproduce - example code ```python3 import cv2 a = cv2.imread("ex.png") a.shape ``` - operating system Red Hat Ubi - Docker - architecture (e.g. x86) haswell - opencv-python version opencv-python-headless-4.5.5.64