Repository navigation
Replies: 1 comment
|
Current File Connector indexing does not preserve explicit page or figure-to-caption links. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
While reviewing the PDF File Connector, I noticed a potential limitation in how image-to-text relationships are preserved.
[pdf_image_utils.py](https://github.com/onyx-dot-app/onyx/blob/main/backend/onyx/file_processing/pdf_image_utils.py)extracts images with page information in their filenames (e.g.,page_3_image_Im5.png).[connector.py](https://github.com/onyx-dot-app/onyx/blob/main/backend/onyx/connectors/file/connector.py)ignores the extracted image filename (_img_name), combines the document text into a singleTextSection, and appends images as separateImageSectionobjects.This seems potentially problematic for multimodal RAG, especially technical documents where diagrams need to be interpreted together with surrounding explanations.
Is this an intentional design trade-off, or is there another mechanism downstream that preserves or reconstructs these relationships?
All reactions