st.file_uploader produces undesired results for some pdfs.

Hi everyone!

Something (undesired) is happening to some pdfs when they are stored in memory by st.file_uploader.

I have a streamlit app that allows user to upload documents (docx, pdf, txt) and automatically processes and cleans them.

The upload and processing works just fine, however I noticed that when I am running the EXACT SAME functions line by line but with the pdf simply inputted from my local storage/file path, it produces very different outcomes (usually for the streamlit upload some words are not correctly processed).

Again, the only difference is that, in scenario A the pdf is uploaded via st.file_uploader and in scenario B the pdf is given by a local path. I therefore assume that the pdf is somehow differently stored by st.file_uploader and I am not sure how to fix this. Please note that the correct output is coming from defining local file paths. The streamlit output ist faulty.

Expected behavior:

Pdf uploaded by defining local file path: 2nd paragraph (correctly processed words):

Actual behavior:

Pdf uploaded over st.file_uploader: 2nd paragraph (words not correctly processed):

Additional information:

This behaviour has only been showing up for pdfs, not for docx, txt files so far.

[willhuang1997]

@kajarenc , would you happen to know what is wrong with this off the top of your head? I looked at some of the code but couldn’t find anything that would help point me in the right direction.

[jonas-nothnagel]

Dear @kajarenc and @willhuang1997

After some further digging, I found the problem to be already mentioned quite a lot by the community. Please see here: https://github.com/streamlit/streamlit/issues/904

I am now closing this now. Thank you very much for your engagement and please excuse the mistake on my side. I hope for the issue above we will be able to find a solution soon.

A temporary fix is to store the file in a temporary folder as outlined here https://github.com/deepset-ai/haystack/issues/2824

file = st.file_uploader("File upload", type=["pdf"])

with tempfile.NamedTemporaryFile(mode="wb") as temp:
    bytes_data = files.getvalue()
    temp.write(bytes_data)
    print(temp.name)