It's not supported right now, but you could always write your own loader for it, and create your own
ImageDocument
instances.
Right now there are two options for images in llama-index
- Generating a caption for the image and retrieving based on the caption
- Applying ocr on the image and retrieving based on the OCR
PyPDF can extract images it looks like, so this could be added to the loader in the future
https://pypdf.readthedocs.io/en/stable/user/extract-images.html