Sci Figure

Sci-Figure — Scientific Figure Extractor

Precisely extract figures and sub-figures from academic PDF papers.

License note: sci-figure is licensed under AGPL-3.0-or-later because it links PyMuPDF (fitz), which is AGPL-licensed.

Installation

Install the package from the skill directory before first use:

bashCopy code
cd ${SKILL_DIR}
pip install -e .

This registers the sh-sci-fig CLI command. Requires Tesseract OCR:

• Windows: winget install UB-Mannheim.TesseractOCR
• Linux: apt install tesseract-ocr
• macOS: brew install tesseract

Preferences (EXTEND.md)

Use Bash to check EXTEND.md existence (priority order):

bashCopy code
# Check project-level first
test -f .baoyu-skills/sci-figure/EXTEND.md && echo "project"

# Then user-level (cross-platform: $HOME works on macOS/Linux/WSL)
test -f "$HOME/.baoyu-skills/sci-figure/EXTEND.md" && echo "user"

EXTEND.md Supports: Default DPI | Default output format | Tesseract path

Usage

bashCopy code
sh-sci-fig <input.pdf> [options]

Options

Examples

bashCopy code
# Extract Figure 2, sub-figure c
sh-sci-fig paper.pdf -f 2 -s c

# Extract entire Figure 3
sh-sci-fig paper.pdf -f 3

# List all available figures in a PDF
sh-sci-fig paper.pdf --list

# Extract all figures
sh-sci-fig paper.pdf --all

# Custom output directory and DPI
sh-sci-fig paper.pdf -f 2 -s c -o ./output/ -d 300

# Use EasyOCR for sub-figure label detection
sh-sci-fig paper.pdf --all --ocr easyocr

# CV-only strategy (skip native extraction)
sh-sci-fig paper.pdf --all --strategy cv

# Render page with annotated bounding boxes (debugging)
sh-sci-fig paper.pdf -f 1 --render-page --annotate

# Manual bbox extraction (multimodal correction)
sh-sci-fig paper.pdf -f 1 --bbox 100,200,800,1200

Output:

Copy code
Extracted: figure_2c.png (1920x1080, 600 DPI)

Error Handling

Tech Stack

Extraction Engines (v2)

The hybrid strategy (default) tries all three in order and validates results.

Detected Figure Fields

Each figure returned by FigureExtractor.detect_all() is a dict with these keys:

list_figures() returns the same dicts without the image field.

Extension Support

Custom configurations via EXTEND.md. See Preferences section for paths and supported options.

© License & Copyright

Aut_Sci_Write — Autonomous Scientific Writer

• Author: Shuo Zhao
• License: MIT License
• Copyright: © 2026 Shuo Zhao. All rights reserved.
• Original Work: This is an original work created by the author. No reproduction, redistribution, or commercial use without explicit permission. Permission is hereby granted, free of charge, to any person obtaining a copy of this software... (See the LICENSE file in the root directory for the full MIT terms.)

This skill is part of the Aut_Sci_Write suite. For full license terms, see the LICENSE file in the project root.