Hi everyone,<\/p>\n
I\u2019m currently stuck on a client project where I need to extract structured data (values, labels, etc.) from charts and graphs<\/strong>. Since it\u2019s client data, I cannot use LLM-based solutions (e.g., GPT-4V, Gemini, etc.)<\/strong> due to compliance\/privacy constraints.<\/p>\n So far, I\u2019ve tried:<\/p>\n While they work decently for text regions, they perform poorly on chart data<\/strong> (e.g., bar heights, scatter plots, line graphs).<\/p>\n I\u2019m aware that tools like Ollama models<\/strong> could be used for image \u2192 text, but running them will increase the cost of the instance<\/strong>, so I\u2019d like to explore lighter or open-source alternatives<\/strong> first.<\/p>\n Has anyone worked on a similar chart-to-data extraction<\/strong> pipeline? Are there recommended computer vision approaches, open-source libraries, or model architectures<\/strong> (CNN\/ViT, specialized chart parsers, etc.) that can handle this more robustly?<\/p>\n Any suggestions, research papers, or libraries would be super helpful \ud83d\ude4f<\/p>\n Thanks!<\/p>\n<\/div>\n submitted by \/u\/Fit-Soup9023 <\/a> \n
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
