Measure every chip. Identify every species.

ChipLab turns a flatbed scan of woodchips into per-chip geometry and species predictions — offline, on your desktop. A companion to DART for forest-products researchers, mill QA, and wood-science labs.

Notice · Beta

ChipLab is in active development. Segmentation, measurement, and species predictions are provided as reference-level outputs and may contain errors. Users should independently verify results before publication or decision-critical use.

What It Does

01

Automatic Segmentation

A Mask R-CNN (MMDetection) model finds each chip in the scan. SAM2 lets you correct masks by clicking — human-in-the-loop where it matters.

02

Calibrated Measurement

Length, width, area, aspect ratio, and perimeter in millimetres. Scanner DPI → mm calibration is built in, so geometry is physical, not pixels.

03

Species Classification

Per-chip species prediction with a confidence score, trained on log-level held-out data to avoid leakage between training and evaluation.

How the Pipeline Works

Scan
Flatbed TIFF
Segment
Mask R-CNN
Correct
SAM2 · click to fix
Measure
Geometry in mm
Classify
Species + confidence
Export
CSV / PDF

From Scan to Segmentation

Drag the handle to compare a raw scanner image with ChipLab’s per-chip masks and identifiers.

Raw Scan Segmented

Illustrative representation. Masks and identifiers are stylized for demonstration; real scans and overlays to be supplied.

Output Preview

chip_idlength_mmwidth_mmarea_mm2speciesconf
000124.611.2198.4Q. rubra0.96
000218.39.7132.1A. saccharum0.91
000331.014.5301.7P. strobus0.88
000415.98.196.3Q. rubra0.94
000527.412.8241.0B. nigra0.79
000621.710.4167.5A. saccharum0.85
000729.213.1268.9P. strobus0.90
000817.59.0118.6C. ovata0.72

Species distribution

Q. rubra
31
A. saccharum
23
P. strobus
18
B. nigra
12
C. ovata
8

Sample values shown. Replace with cleared rows from your own run.

Methods

Dataset design

15 species × 3 logs × 1 bag × 3–5 layouts, with a strict log-level train / test split so no chip from a test log appears in training.

Models

Mask R-CNN for segmentation, EfficientNet / ResNet-50 for species classification, and SAM2 for human-in-the-loop mask correction.

Evaluation

Leave-one-log-out per species, reported with confidence, so accuracy reflects generalization to unseen logs rather than memorized chips.

Reproducibility: ChipLab runs fully offline and is deterministic given the same image, calibration, and model weights. Document software version, model weights, and dataset identifiers in any derived work. Full method plan to be linked on release.

Download & Install

Windows 10 / 11
ChipLab-Setup.exe
Download InstallerWIP

Built on Electron + Python. Runs fully offline. No data leaves your machine.

  • Windows 10 / 11, 64-bit
  • ~8 GB RAM
  • Optional CUDA GPU for faster inference

Installer link, version, and release date to be provided via the GitHub releases page.

Technology Stack

Electron / Node.js
Python
Mask R-CNN (MMDetection)
SAM2
EfficientNet / ResNet-50

How to Cite

When ChipLab is used in analysis, exports, or figure generation, cite the tool separately in addition to any dataset citations.

@software{chiplab2026,
  title   = {DART: ChipLab},
  author  = {{Iowa Forestry Media Library}},
  year    = {2026},
  version = {Beta},
  url     = {https://iowaforestrymedialibrary.org/chiplab.html}
}

Companion tools. ChipLab shares calibration and reporting conventions with DART, the tree-ring analysis workbench. Questions? Contact [email protected].