{"id":45742992,"title":"Show HN: UnPlotter – Extract data from vector PDF plots","points":2,"user":"ramcdona","time":1761716040,"time_ago":"17 hours ago","type":"link","content":"

I thought you all might be interested in this recent project of mine.

Here is a 78 second video demonstrating its use: https://youtu.be/AkR6-IWVgBU

UnPlotter is a JS app that will extract the data from vector plots in PDF files. Perhaps you want to compare your method with a published result. Or perhaps you want a MOSFET performance curve from a product data sheet.

Over the years, there have been many programs to extract data from raster plots. They typically use image processing / computer vision algorithms to 'follow the red line' -- and then scale the results to recover an approximation of the data.

When I have needed to extract data from figures that were available as vector files, I was unwilling to take a screenshot and then use a raster extraction tool. Instead, I would delete all the extraneous data in a vector drawing program (say InkScape or Illustrator). Then I would convert to a text-based vector format (usually EPS, but more recently SVG). Then I would edit the file and reverse engineer the curves from the data. It was painstaking and slow, but it satisfies a certain level of OCD.

UnPlotter does all this so you don't have to.

The good news -- no image processing / CV algorithms needed to follow lines. Most plotting programs group the curves in a way that they are intrinsic.

The accuracy is shocking -- although plotting programs may round their output, PDF stores real numbers as single precision floating point (in pt (1/72 in) in page coordinates) -- so some scaling occurs can introduce some truncation error, but you're effectively recovering the data to single precision (say less than 1e-5 error). Which is a ton better than picking pixels from a 300dpi image.

Everything happens in your browser, the loaded PDF file never leaves your computer. No logging, tracking, etc.

I'd always wondered why nobody ever made a tool to do this -- so I did. I hope some of you find it interesting and perhaps even useful.","url":"https://www.unplotter.com/","domain":"unplotter.com","comments":[],"comments_count":0}