I had the same issue with vertical container plates and ISO codes. My solution was to detect the bounding boxes, and then rotate them horizontally before training my OCR model. After preprocessing my dataset this way, my OCR model successfully extracts text from vertical plates. Hope this helps!