Summary

FigMirror automates the creation of publication-quality figures by mirroring the style of reference figures from research papers. It targets researchers and engineers needing to generate consistent, editable plots, significantly reducing manual styling effort and producing camera-ready outputs.

How It Works

The core is an agentic Drawer-Reviewer loop. The Drawer renders figures using "Grounded Measurement," which leverages pixel-level analysis (Measurement with Axis, Resonate with Code) for precise visual target identification and executable checks. The Reviewer critiques the output against a reference, guiding iterative refinement. For 3D plots, geometry-aware prompting preserves complex compositions. An Aesthetic Lib provides fallback rules. This approach yields editable matplotlib scripts and ensures style consistency.

Quick Start & Requirements

• Installation: Clone the repo (git clone https://github.com/VILA-Lab/FigMirror.git). For the Web UI, run bash scripts/install.sh followed by uv run python scripts/figcopy_serve.py (requires uv, install via python3 -m pip install uv). For agent integration (skill), use curl -fsSL https://raw.githubusercontent.com/VILA-Lab/FigMirror/main/scripts/install.sh | bash.
• Prerequisites: Requires AI backends like Claude or Codex. Python 3.x environment recommended.
• Links: Web UI runs at http://127.0.0.1:8765/. Detailed methods are in docs/method.md.

Highlighted Details

• Offers a gallery of 139 pre-defined figures across 25 chart types for style inspiration.
• Outputs editable matplotlib scripts and camera-ready PDFs.
• Specialized handling for 3D figures, including geometry-aware prompting.
• Grounded Measurement enables fine-grained, pixel-based visual analysis and validation.

Maintenance & Community

• Actively seeks contributions for bug fixes, showcase examples, prompt improvements, and UI polish.
• Roadmap includes developing automated benchmarking tools and curated datasets.
• No community channels (e.g., Discord, Slack) are listed.

Licensing & Compatibility

• The repository README does not specify a software license. This lack of clarity may impact commercial use and integration decisions.

Limitations & Caveats

• Relies on specific AI model backends (Claude/Codex).
• The setup process may require troubleshooting.
• Absence of a stated license is a significant adoption barrier.
• Benchmarking and evaluation protocols are still under development.