Image and data visualization distinction

[renesteeman]

Is it possible to differentiate between an image, such as a photo, logo, or face, and a visual representation of data, such as a bar chart, line graph, or diagram?

[JulioZhao97]

@renesteeman Could you please provide more explaination? I did not follow your question

[renesteeman]

@JulioZhao97 Sure. So for my use case, I need a way to distinguish between two types of figures. The first are those that present some sort of useful information to extract, such as graphs and diagrams. The second type would consist of company logos, decorative pictures on slides, pictures of a person, etc. that do not contribute any usable information. This is mainly for corporate documents such as investor presentations. The goal is to get the page numbers for where graphs or diagrams can be found so that their contents can be analyzed without having to look over entire files.

So the question is if it is possible to distinguish the output class of 'figure' into a sub-class of 'data visualization' and 'images'.

[JulioZhao97]

@renesteeman Sorry for late reply (Chinese New Year). For now we do not distinguish images into figure/diagram/logo or so on. I suggest:

1. Train your own model on dataset that fits your requirements. Or
2. Using a classifier or LVLM to distinguish detected images into diagram or logo and so on.