Data and Supplementary Information for <b><i>"Leveraging Vision Capabilities of Multimodal LLMs for Automated Data Extraction from Plots"</i></b>

Data and Supplementary Information for <b><i>"Leveraging Vision Capabilities of Multimodal LLMs for Automated Data Extraction from Plots"</i></b><br><br>https://arxiv.org/abs/2503.12326<br><br>Within the datasets, files ending in "out_data" contain the results of LLM vision data extraction.<br>Files ending in "out_code" contain the LLM generated core used to recreate the source plot using the extracted data<br>Files ending in "out_conversation" contain the entire conversation with the LLM, as conducted by PlotExtract.<br>Files starting with "interpolated" show the interpolation comparison for each set on each plot, as described in the main text of the paper.<br>Files starting with "pointwise" show the pointwise comparison for each set on each plot, as described in the main text of the paper.<br>Accuracy statistics are present both in the comparison images and in files ending in "stats".<br>The files ending in "csv" contain the manually extracted ground truth.<br>All of these files are generated as PlotExtract is executed.<br><br>The published dataset does not contain the source images, only all of the outputs, due to copyright. Instead all images used are listed in the included spreadsheet with the DOIs of the source document and the Figure number so that they are easily identifiable.<br>All of the codes, including PlotExtract and codes used for evaluation and comparisons are located in MPPolak_DMorgan_PlotExtract_Codes.zip" and they include instructions on usage and should be self-explanatory.