This is a small script written by DonMischo which I modified slightly to make it a bit more flexible. I added support for multiple image-formats, and it now asks you for the model name when running the script instead of having to edit the script manually each time.

Place the script in a folder, and your dataset of images sorted by one folder per concept, and those folders should all be in the same folder as the script. Like this:

Then run the script, and enter the name of your model. This name will be used as the prefix for the captioning. It should then generate a caption-file for each image, placed inside the folder for the concept. It will use the folder name for the concept as part of the caption.