Quality of Screenshots #345

Open
opened 2026-02-16 00:17:34 -05:00 by yindo · 4 comments
Owner

Originally created by @davidrzs on GitHub (Nov 20, 2024).

Originally assigned to: @hexapode on GitHub.

Is there an option to control the quality of the screenshots? Ideally we would like to show a preview of the page where the text was extracted, but with the given resolution this is not possible as it looks very pixelated.

Is there an option to select a higher resolution for the screenshots?

Originally created by @davidrzs on GitHub (Nov 20, 2024). Originally assigned to: @hexapode on GitHub. Is there an option to control the quality of the screenshots? Ideally we would like to show a preview of the page where the text was extracted, but with the given resolution this is not possible as it looks very pixelated. Is there an option to select a higher resolution for the screenshots?
yindo added the enhancement label 2026-02-16 00:17:34 -05:00
Author
Owner

@hexapode commented on GitHub (Nov 21, 2024):

Will see if possible, may incur extra cost.

@hexapode commented on GitHub (Nov 21, 2024): Will see if possible, may incur extra cost.
Author
Owner

@BinaryBrain commented on GitHub (Nov 21, 2024):

Hi @davidrzs,
What would be an ideal resolution for your use case?

@BinaryBrain commented on GitHub (Nov 21, 2024): Hi @davidrzs, What would be an ideal resolution for your use case?
Author
Owner

@davidrzs commented on GitHub (Nov 21, 2024):

Our use case is previewing from where a chunk comes.

One idea would be to have a PDF where we jump to the correct page. The problem is that this doesn't work with word files as you might render it differently from how we would transform it into a PDF (e.g. if we do it with soffice we likely have the layout different from you), and extracting from a word is easier than to directly extract from the PDF (correct me if I am wrong).

So the "cheap" alternative would be to display the image of the rendered page instead of the PDF version with an embedded PDF.js. But for this to work the resultion must be great enough such that a normal A4/letter sized page is readable (which is not necessarily the case now).

Let me know what you think would be the best way forward - not sure what the best solution to this use case is.

@davidrzs commented on GitHub (Nov 21, 2024): Our use case is previewing from where a chunk comes. One idea would be to have a PDF where we jump to the correct page. The problem is that this doesn't work with word files as you might render it differently from how we would transform it into a PDF (e.g. if we do it with soffice we likely have the layout different from you), and extracting from a word is easier than to directly extract from the PDF (correct me if I am wrong). So the "cheap" alternative would be to display the image of the rendered page instead of the PDF version with an embedded PDF.js. But for this to work the resultion must be great enough such that a normal A4/letter sized page is readable (which is not necessarily the case now). Let me know what you think would be the best way forward - not sure what the best solution to this use case is.
Author
Owner

@davidrzs commented on GitHub (Dec 2, 2024):

Sorry for pinging, but wondering if you have some thoughts on the use-case above, i.e. which direction you would see most feasible. (we will likely decide internally what to do / how to do that in-house later this week).

@davidrzs commented on GitHub (Dec 2, 2024): Sorry for pinging, but wondering if you have some thoughts on the use-case above, i.e. which direction you would see most feasible. (we will likely decide internally what to do / how to do that in-house later this week).
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: run-llama/llama_cloud_services#345