WebPlotDigitizer

WebPlotDigitizer's core value is that it does one job and does it broadly. Where a lot of chart-reading tools handle only XY line plots, this one covers XY (line and scatter), bar charts, histograms, polar diagrams, ternary diagrams, and circular chart recorders — plus simple measurements on images that aren't charts at all. That breadth is why it shows up in thousands of published studies and why v5 layers multimodal machine learning methods on top of the older computer vision algorithms rather than replacing them. The workflow is honest about its limits. You calibrate the axes, including logarithmic and date-time axes, then choose between automated detection (the Wand tool and the newer AI assist features) and manual clicking. When the automation fails on a noisy, low-contrast, or overlapping figure, you fall back to placing points by hand — and the tool gives you undo/redo, keyboard shortcuts, zoom and pan, background filtering, and color inversion to make that tolerable. You can average repeated extractions to reduce error, validate visually through the Plotly integration, and export to CSV or JSON or copy straight to the clipboard. The practical constraints matter. It is designed for one-off images rather than high-throughput pipelines: you import each image and export each result yourself, there is no batch processing, and there is no API for programmatic access. It is not a charting or visualization tool, it does not stream data, and it does not OCR text labels or legends — it extracts numerical series. The interface is functional rather than polished, and there are no collaboration features, so a team working through the same figures will pass files around rather than work in a shared workspace. The desktop build matters if you need to work offline. Where it fits: a researcher digitizing a dose-response curve from a pharmacology paper, an engineer recovering a torque-vs-rpm curve from a datasheet, a student rebuilding a textbook figure as a spreadsheet, or anyone with a scanned report and no original data file. Where it doesn't: anything you need to do a thousand times unattended, anything requiring text recognition, and anything where the end goal is a published interactive chart rather than a numbers table.