Different behaviour when feeding different types of input

[beng-ern]

I was running test on a few sample images and I have this one particular image that runs into OpenCV error of "error: (-215:Assertion failed) !ssize.empty() in function 'cv::resize'" when I feed it from the path directly into easyocr.Reader.readtext, instead of first converting the photo to numpy array using cv2.imread.

Upon some investigation, I found that it is because of width of one of the boxes detected is negative (x_max is smaller than x_min at 'get_image_list' function of utils.py file).

However, when I first convert the photo into the numpy array as an input to easyocr.Reader.readtext, the same error did not occur.
I have tried to compare both input types (image file/numpy array) of the other photos, and none of them run into the same error...

So I am just wondering is it a normal behaviour that different input types generating different detection boxes? If it is normal, should there be some further checking added on the box width...?