OpenAI-Captioner

Uses GPT models from OpenAI to caption image data in single & batch configurations.

3 important steps

1. make an openai account in order to setup an api key (side note: openai will mandate that the user sets up a CC with their own respective account to manage the billing for their api access/usage)
2. clone the repo
3. pip install -r requirements.txt

Usage

There are textbox entries for the user to provide information for the following:

• api key
• prompt
• output PATH
• preset configuration name : for preset dropdown list

Additional Features

• image and caption/tag statistics tab, for evaluating as a (post-processing step) all the results from the model
• preset options for the prompt & output path
• new queue system to let the user provide as many images as they want to avoid the 20 image limit with the api calls
• images that might not return captions can temporarily stay on the image upload component, to quickly test other GPT vlm models available to see if those provide captions instead
• a second image upload component was added to let the user add images to the live queue as they deem necessary

Run Instructions

python gpt-vlm-run.py

Main Menu

Batch Mode Configuration Tab

Stats Tab

Troubleshooting

• if the response from the api call contains something along the lines of: "Sorry image could not be described", then the image is likely to in some way be in violation of OpenAI's T.O.S.
• if the response eludes to a image incompatibility; could be the result of the selected model being deprecated and only raw image urls would be supported (versus the current image string conversion)
• if other errors occur, see: https://platform.openai.com/docs/guides/error-codes/api-errors.