YS.Yuming Su 苏禹铭 中文
Home / All software / GBLine
Spectral digitization · Evidence-guided review

GBLine

Research prototype

A local workflow connecting OCR, coordinate calibration and curve extraction with optional vision-language-model review for scientific charts and chemical spectra.

PythonOCRMCP
Discuss this software ↗Online compute demo

Email verification and administrator approval required. Shared per-user limits: 1 running task, 2 queued tasks and 10 submissions per day; private results are retained for 72 hours. Service availability depends on the research server.

Workflow: image input, axis and curve extraction, optional evidence review, numerical output and diagnostics.
Workflow schematic. Model review is optional; numerical quality still requires checking.

From a spectrum image to numerical curves

Online calculation

Use Online compute demo above, register with an email and password, verify your email and wait for administrator approval. In GBLine, upload a PNG or JPEG chart (up to 10 MiB), select the curve and point counts and submit. The server runs a fixed extraction workflow without external model calls. Review the calibration preview and diagnostics before downloading the curves. The online version supports one panel with one X/Y axis pair; uncertain extraction is flagged for review.

Each task is limited to 2 CPU threads, 4 GiB RAM and 5 minutes. All tools share 1 running task, 2 queued tasks and 10 daily submissions per user. Private results expire after 72 hours.

GBLine is a local tool for digitizing scientific line charts and chemical spectra. It connects OCR, coordinate calibration and pixel-based curve extraction with an optional vision-language-model review workflow through MCP.

The numerical core performs coordinate conversion and curve tracing. A visual model can inspect visible tick labels and curve colors, propose corrections and respond to the tool's diagnostic feedback.

Main workflow

  • Start with a single-panel chart and inspect the plotting area, axes and visible ticks.
  • Establish a linear or log10 coordinate mapping from image evidence.
  • Extract curve pixels; review ambiguous axes, missing curves or colors when requested.
  • Retrieve full numerical curves and diagnostic records, then check coverage and numerical fidelity before reusing the data.

The MCP interface includes inspect_chart_axes, extract_chart_data and get_curve_full_data. Depending on the available evidence, the tool can return extracted data, request further review or abstain.

What to check in the output

Correct axis calibration and the expected number of curves do not guarantee a complete trace. Inspect horizontal-domain coverage, gaps and the extracted curve against the source image. Multi-panel figures should be separated first; dual-y-axis plots and insufficient visible calibration evidence are outside the current workflow.

The research currently includes developmental and retrospective validation; independent evaluation on unseen spectra remains ongoing.

Access and research status

The current local MCP implementation is documented as version 2.0.0. Model assistance requires a compatible image-capable agent and its selected provider; the numerical tool itself is not tied to a particular model service.

Contact Yuming Su for the current software and setup. A revised research manuscript is in preparation.