Line chart with data markers
11 markers on a curve. All 11 values were read within 0.1% of the y-axis range.
AI reads the axes and legend · Images & PDFs · CSV & Excel export
AI plot digitizer: extract data from graph images
Upload a research paper PDF or a chart image and this online AI plot digitizer extracts the data points as a table, with no axis calibration.
No chart at hand? Open a sample:
Your chart image is sent to our AI model to be read and is not stored. A PDF stays in your browser; only the chart you select from it is sent. The extracted numbers are saved to your history.
What is a plot digitizer?
A plot digitizer is a tool that turns the image of a graph back into numbers. A plot digitizer, also called a graph digitizer or chart data extraction tool, reads the axis scales and the position of every point, bar or curve, and returns x–y values you can open in Excel, R or Python. Researchers use plot digitizers to extract numerical data from figures in published papers when the raw data is not available.
How to extract data from a graph image
Three steps, no axis calibration. Reading one chart usually takes 10–30 seconds.
1
Upload a chart image or a PDF
Drop in a PNG, JPG or WebP, or paste a screenshot with Ctrl/⌘ + V. For a figure in a paper, upload the PDF, go to the page and drag a box around the chart. Use one chart at a time.
2
AI reads the axes, legend and data
The AI plot digitizer reads the tick labels, decides whether each axis is linear, logarithmic, dates or categories, takes the series names from the legend and returns every series as numbers.
3
Check, fix and export
The extracted data is re-plotted next to your chart. If the two plots match, the numbers are right. Edit any cell that is off, then copy the table or download CSV or Excel.
Sample charts and the data read from them
Each chart below was drawn from data we generated, then read by the AI plot digitizer. The plot on the right is drawn from the extracted numbers, unedited.
Logarithmic y axis, two series
Two decay curves on a log scale. The legend names became the two column headers.
Grouped bar chart
Five groups with two bars each. Category labels became the first column.
Error bars on a low-resolution JPEG
A 572-pixel-wide JPEG. Each error bar came back as a low and a high column.
How accurate is the AI plot digitizer?
On six test charts with known values, the largest error on charts with visible markers or bars was 0.31% of the y-axis range. Smooth curves without markers were less exact: the largest single-point error was 4.04%.
| Test chart | Data points | Mean error | Largest error |
|---|---|---|---|
| Line chart with data markers | 11 | 0.03% of y-axis range | 0.1% of y-axis range |
| Grouped bar chart, 2 series | 10 | 0.09% of y-axis range | 0.22% of y-axis range |
| Low-resolution JPEG with error bars | 7 | 0.13% of y-axis range | 0.31% of y-axis range |
| Scatter plot, 2 groups with overlapping markers | 24 | 0.15% of y-axis range | 0.31% of y-axis range |
| Logarithmic y axis, 2 series | 18 | 0.96% of the value | 1.41% of the value |
| 3 smooth curves without markers | 47–52 | 0.82% of y-axis range | 4.04% of y-axis range |
How this was tested
- Method: we drew six charts with matplotlib from data we generated, so the true value of every point is known. The AI plot digitizer read each chart three times on November 11, 2026, 18 readings in total. For each chart the table shows the worst of the three runs.
- Error measure: on linear axes, the difference between the extracted and the true value divided by the y-axis data range. On the logarithmic axis, the difference divided by the true value. For curves without markers, each extracted point is compared with the curve's formula at the same x.
- Harder images, one or two runs each: a noisy, slightly rotated grayscale JPEG that imitates a scanned figure had a mean error of about 2% and a largest error of about 4% on 16 points. A scatter plot with 60 points returned all 60, within 0.5%. A 420-pixel-wide chart with four lines returned all 24 points exactly.
- Limits: all test charts were drawn by computer, including the imitation scan. Real scans, photos of slides, very small text and heavily overlapping series can be less accurate. The numbers come from an AI model reading the image, not from pixel measurement, so always compare the re-plotted chart with the original before you use the data.
- The scatter test found all 24 points in every run. Error-bar half-lengths were read as 1.7–1.9 units where the true value was 1.8.
Tested by the GAAbstract team. About GAAbstract
Which charts can the plot digitizer read?
The AI plot digitizer works on 2D charts with labelled axes. The list below is what we have tested, not a promise for every figure.
Measured in the accuracy test
- Line charts with data markers
- Smooth curves without markers
- Scatter plots with several groups
- Grouped bar charts
- Logarithmic axes
- Error bars
Read correctly in a single test
- Date and category axes
- Histograms (one count per bin)
- Box plots (min, Q1, median, Q3, max per box)
- Black-and-white lines told apart by dash style
- Stacked bar charts (one value per segment)
- Two y axes on one chart
- Pie charts (percent per slice)
Not supported
- Polar and ternary plots
- Maps, heat maps and contour plots
- 3D surface plots
- Several panels in one image — crop to one chart first
- Password-protected PDFs — paste a screenshot of the figure instead
The re-plotted check chart draws stacked bars side by side and puts both y axes on one scale, so for those two chart types compare the table values rather than the shape.
From figure to spreadsheet
What the AI plot digitizer does for you
Manual plot digitizers make you calibrate the axes and trace each series. The AI plot digitizer does those steps and leaves you the checking.
Axes read automatically
Tick labels, units and scale type are read from the image. Linear, logarithmic, date and category axes need no setup and no calibration points, and grid lines do not have to be masked out.
One column per legend entry
Series names come from the legend. A chart with three curves returns three named columns, not one mixed list of points.
Error bars as low and high values
When points carry error bars, the lower and upper ends are returned as two extra columns, ready for a meta-analysis spreadsheet.
A re-plot to check against
The extracted numbers are drawn as a chart next to your image. A wrong point shows up as a shape that does not match.
Editable table
Click any cell to correct a value or delete a row. The check chart redraws as you edit, and exports use the edited table.
CSV, Excel and clipboard export
Download a .csv or a real .xlsx file with numeric cells, or copy the table and paste it into Excel or Google Sheets. Choose side-by-side columns or a long tidy layout for R and Python.
AI plot digitizer vs. manual plot digitizers
Manual tools give you pixel-level control but need calibration clicks for every image. Facts about other tools were checked on their own pages on November 11, 2026.
| Axis calibration | GAAbstract AI Plot Digitizer | WebPlotDigitizer 4.8 (archived free version) | PlotDigitizer (free online app) | graph2table |
|---|---|---|---|---|
| Axis calibration | Automatic — read from the tick labels | Manual — click known points on each axis and type their values | Manual — place four calibration points and enter their values | Automatic, with draggable axis anchors |
| Several series | Named from the legend, one column each | Each dataset is extracted separately, picked by colour | Multiple datasets are a Pro feature | Automatic |
| Polar, ternary and map plots | Not supported | Supported | Supported | Not stated |
| Export | CSV, Excel (.xlsx), clipboard | CSV, clipboard | CSV, Excel, JSON, clipboard | CSV, Excel (.xlsx), clipboard |
| Account | Not needed for the trial | Not needed | Not needed | Needed to see the data table |
| Price | 1 credit per chart; trial credits without an account | Free | Free with limited features; Pro licence $40 one-time (promotional price, list price $80) | 2 free conversions on sign-up; packs from $7.90 for 10 |
Sources: WebPlotDigitizer 4.8, PlotDigitizer, graph2table pricing, graph2table docs, WebPlotDigitizer 5 release notes
Which plot digitizer is best for your chart?
For standard 2D charts with labelled axes, an AI plot digitizer is the fastest choice, because it reads the axes and every series in one step. Choose a manual tool such as WebPlotDigitizer 4.8 or PlotDigitizer when the chart is a polar plot, a ternary diagram or a map, when you need to place each point yourself at pixel level, or when you do not want an AI model to process the image. Manual tools take longer per chart but do not depend on an AI model reading the image correctly.
What researchers use a plot digitizer for
Meta-analysis
Pull group means and error bars out of published figures when a paper reports them only as a chart.
Comparing with published results
Recover the curve from a figure in a paper to overlay it on your own data when the authors did not share the raw numbers.
Datasheet and reference curves
Turn a pump curve, a stress–strain diagram or a calibration curve from a datasheet into values you can interpolate.
Graph to Excel
Rebuild a chart from a report or a slide as an editable Excel chart by extracting its data first.
How to convert a graph image to Excel data
To convert a graph to Excel data, extract the values from the image first, then open them as a spreadsheet.
- Upload the graph image, or the PDF that contains it, to the plot digitizer above and click Extract data.
- Compare the re-plotted chart with your graph and correct any cell that is off.
- Click Download Excel (.xlsx) for a file with numeric cells, or Copy for Excel / Sheets and paste into an open workbook.
- In Excel, select the columns and insert a Scatter or Line chart to redraw the graph.
Plot digitizer FAQ
Common questions
Turn your next figure into data
Upload a chart, compare the re-plot with the original and download the table as CSV or Excel.