Overview

I recently created the following article and notebook.

At the time of writing the above article, only the following input format was supported.

Image file mode (specified with -s f) (Use this when providing a single image file as input)

However, through verification in the following article, it became clear that applying the above option to multiple images incurs significant overhead.

Therefore, I modified the notebook to also support the following input format.

Single input dir mode (specified with -s s) *Default

Below, I explain how to use the above input option.

Usage

In “2. Settings,” select “s” for “input_structure.” To use the conventional “Image file mode,” select “f.”

When “s” is selected for “input_structure,” the “extensions” field is ignored.

Also, there is an important note about how to create the input folder. Let us take the following input folder path as an example.

/content/drive/MyDrive/ndl_ocr/input/

In this case, create an “img” folder under the “input” folder and place the images inside it, as shown below.

After completing the above preparation and settings, you can execute OCR processing using “Single input dir mode.”

Summary

Regarding the notebook created this time, when “f (Image file mode)” is selected, you can run it without being particularly conscious of the directory structure within the specified input folder. However, in this mode, the program runs for each image file, which incurs some overhead. When processing a large number of image files, it is recommended to select “s (Single input dir mode).”

There may be points where my understanding is incorrect, but I hope this is helpful.