figpanel

Detect and extract individual panels from scientific figures using YOLOv12.

Scientific papers often combine multiple plots into a single composite figure (Figure 1a, 1b, 1c...). figpanel automatically detects these sub-panels, reads their labels via OCR, and extracts each one as a separate image.

Quick Start

pip install figpanel

import figpanel

# Detect subplots and captions (bounding boxes + confidence scores)
results = figpanel.detect("figure.png")
# {'subplots': [(x1,y1,x2,y2,conf), ...], 'captions': [...]}

# Visualize detections on the image
figpanel.visualize("figure.png", save="annotated.png")

For the full pipeline (detect + OCR + match + crop + deduplicate):

pip install figpanel[full]

# Extract individual panels
panels = figpanel.extract("figure.png", "output/")

for panel in panels:
    print(f"Panel {panel.label}: bbox={panel.bbox}, confidence={panel.confidence:.2f}")
    panel.image.show()

Examples

Blue boxes = detected subplots, Green boxes = detected captions. Images from open-access papers (CC BY 4.0).

Features

• YOLOv12 Detection - Trained on 5,000+ annotated scientific figures to detect subplots and caption labels
• Automatic Model Download - Model weights are hosted on HuggingFace and downloaded automatically on first use
• OCR Caption Reading - Reads single-character panel labels (a, b, c...) using Tesseract
• Smart Matching - Greedy nearest-neighbor algorithm matches captions to their corresponding subplots
• Deduplication - Removes duplicate panels using perceptual hashing (pHash) and ORB feature matching
• Two-Tier Install - Base install for detection only; [full] adds OCR, matching, and deduplication

Model Details

Installation Options

# Detection only (ultralytics + huggingface_hub)
pip install figpanel

# Full pipeline (+ pytesseract, imagehash, opencv)
pip install figpanel[full]

Note: The [full] install requires Tesseract OCR to be installed on your system.

API Reference

figpanel.detect(image_path, conf=0.25, iou=0.5)

Run YOLO detection on a scientific figure.

Returns: dict with "subplots" and "captions" keys, each containing a list of (x1, y1, x2, y2, confidence) tuples.

figpanel.extract(image_path, output_dir=None, *, conf=0.25, iou=0.5, padding=4, dedup_thresh=0.3, prefix=None)

Full pipeline: detect, OCR, match, crop, and deduplicate panels.

Returns: list[Panel] where each Panel has .label, .bbox, .confidence, and .image attributes.

Requires: pip install figpanel[full]

figpanel.visualize(image_path, save=None, *, conf=0.25, iou=0.5, line_width=3)

Draw detection boxes on the image. Subplots in blue, captions in green.

Returns: Annotated PIL.Image.

Use Cases

• Literature Mining - Automatically decompose composite figures for indexing and search
• Dataset Creation - Build plot-type datasets from scientific papers at scale
• Accessibility - Extract individual panels for screen readers and alternative text
• Research Tools - Integrate panel detection into paper analysis pipelines
• Quality Control - Validate figure composition in manuscript preparation

About Plottie

figpanel is built and maintained by Plottie - the scientific plot discovery platform. Plottie helps researchers explore, collect, and find inspiration from high-quality scientific plots across open-access literature.

Visit plottie.art to browse thousands of curated scientific figures.

Citation

If you use figpanel in your research, please cite:

@software{figpanel,
  title = {figpanel: Scientific Figure Panel Detector},
  author = {Plottie},
  url = {https://github.com/Plottie/figpanel},
  version = {0.1.0},
  year = {2026},
  note = {Built by Plottie (https://plottie.art)}
}

License

This project is licensed under the AGPL-3.0 license. The YOLOv12 model weights inherit this license from Ultralytics.

Academic and research use is unaffected. If you integrate figpanel into a network service, you must make your source code available under AGPL-3.0.