Output dir doesn't work for extracted images

[atmoraes1]

I'm trying to save the extracted images to a dir but it seems two parameters are not working:

1. extract_image_block_to_payload: if set to false the image base64 still shows up in the response
2. extract_image_block_output_dir: I think it was not mapped.

I think that's an important step offloading all data to a folder instead of putting it all in the response.

[christinestraub]

@atmoraes1 Can you provide the code snippet you're trying?

[atmoraes1]

It is a simple request using Postman.

The cURL is as follows:

curl --location 'http://localhost:8000/general/v0/general' \
--form 'files=@"<path-to-file>"' \
--form 'coordinates="false"' \
--form 'languages="por"' \
--form 'pdf_infer_table_structure="true"' \
--form 'extract_image_block_types="[\"Image\"]"' \
--form 'strategy="hi_res"' \
--form 'extract_image_block_to_payload="false"' \
--form 'extract_image_block_output_dir="/tmp"'

I expected the image files to be stored in the /tmp dir, but they don't get saved, instead they are returned as base64 in the response.

[christinestraub]

Yes, it is not supported to store image files when using the API. You'll need to use the "unstructured" library directly in your code to store image files.

[atmoraes1]

IMHO it doesn't make sense to build a shim just to save images to an output dir as we already have unstructured as an API.

Can I proceed with a new PR for that?

[awalker4]

Unfortunately that doesn't quite fit with our design of the server. In most cases, the client making the request isn't on the same host, and so passing an output_dir wouldn't provide access to the files. What we can do here is update the partition_via_api function to make the remote partition call, and then pull the images back out of the response. I'll close this and make a new issue.