Horizontal text images's bounding box positions are not set properly.

[maeriil]

In the file imagetranslation.py, for each merged section, we set the start_pos and end_pos to crop out the text. This is what it looks like currently:

        start_pos = []
        end_pos = []

        if should_rotate: # this is true when language is part of vertical text
            start_pos = [
                section.start_pos[1],
                orig_img_width - section.end_pos[0],
            ]
            end_pos = [
                section.end_pos[1],
                orig_img_width - section.start_pos[0],
            ]

        cropped_section = crop_image(image, start_pos, end_pos)

However, if the source language is not part of vertical text, the start_pos and end_pos aren't set, for example if source language is korean. We want to fix this issue.