Understanding speed of Tesserocr's image_to_text

[parulsingh23]

The ReadMe file states that:

image_to_text can be used with threading to concurrently process multiple images which is highly efficient.

However, I'm curious as to how much faster this is. For example, if I were to run tesseract on 120 images each around 100x30 pixels, the average time is .18 seconds per image.

How would running Tesserocr's image_to_text on 120 images each around 100x30 pixels (all in a thread) take?
Additionally, how would this time compare using a computer's CPU, versus a GPU (like provided on google collab, or AWS EC2 instances)?

[ichenjia]

You shouldn't use image_to_text if you have multiple images. Load model and establish the API takes time. You are better off doing something like this:

tess_api=tesserocr.PyTessBaseAPI()

for img in imgs:
  tess_api.SetImage(img)
  text = tess_api.GetUTF8Text()

tess_api.End()