A new tool called LinePilot Digitizer can turn a picture of a line chart back into the numbers that made it.
The tool, built by a group of researchers, reconstructs data series from line plots by tracking curve colors and calibrating the x and y axes. It comes in three versions: a standard manual mode, an enhanced mode, and an OCR mode that reads axis labels automatically using optical character recognition. The team also built DigitizerBench, the first dedicated benchmark for grading these tools, testing them across different chart styles, signal types, and layouts. On the full version of that benchmark, the OCR mode topped the other automatic pipelines with a mean error score of 0.672 and 38.2% of its outputs rated trustworthy enough to use as-is; on a smaller, easier "lite" version, the enhanced mode did far better, with a 0.081 error score, a 100% success rate, and 93.3% trusted usability.
Charts are where a lot of research and business data effectively gets stuck, locked in pixels once a paper or slide deck ships, with no attached spreadsheet. A tool that can reliably read numbers back out of an old chart would save researchers, journalists, and analysts from squinting at gridlines or emailing authors who may not have the original file anymore. The benchmark may matter as much as the tool itself, since there has not been a standard way to compare digitizers' accuracy claims until now.
The math is the reality check here: on the harder benchmark, even the best automatic mode produced usable output well under half the time, so this is a promising research prototype, not a plug-and-play data-recovery button yet.