HERZFELD (Core)

High‑fidelity Epigraphic Rendering for Zonated Feature Extraction and Labelled Datasets HERZFELD is the high-performance numerical engine of the broader HERZFELD Python Suite. It is a specialised library written in Rust designed to process large-scale synthetic datasets for the Optical Character Recognition (OCR) of ancient inscriptions, specifically Middle Persian and Inscriptional Pahlavi.

🦀 Why Rust for Epigraphy?

While the synthetic data is generated within Blender, the post-processing of multi-layered high-dynamic-range (HDR) tensors requires significant computational throughput. The Rust core provides:

• Memory Safety: Guaranteed safety when parsing complex OpenEXR byte streams.
• SIMD Acceleration: Optimised topographic normalisation and relief calculations.
• Parallelism: Utilises Rayon to process thousands of “takes” (render outputs) across all CPU cores.
• PyO3 Integration: Seamlessly exposed as a high-level Python module for researchers.

✨ Key Capabilities

• Adaptive Topographic Normalisation: Automatically scales Z-buffer data to the micron-level relief of a specific inscription (e.g. isolating a 176.7mm relief from a 1.6m camera distance).
• Multi-Layered EXR Parsing: Efficient extraction of Depth, Normal, and Semantic Mask channels from 32-bit float OpenEXR files.
• Zero-Copy Interoperability: Uses numpy-rust to share memory between the Rust engine and Python’s ML ecosystem (PyTorch/TensorFlow) without overhead.

⚙️ The Pipeline Architecture

HERZFELD operates as a hybrid engine:

1. Generation (Blender Core): A procedural simulation environment using Geometry Nodes to synthesise 3D geometry.
2. Extraction (Rust Backend): This crate. A dedicated high-throughput suite that performs adaptive depth normalisation and automated semantic mask generation.

🛠 Usage

For Rust Developers

Add this to your Cargo.toml:

[dependencies]
herzfeld = "0.1.0"

For Python Researchers (Recommended)

This crate is primarily intended to be consumed via the herzfeld Python package. To install the pre-compiled binaries:

pip install herzfeld

📌 Status & Roadmap

This project is currently in active early-stage development (Alpha).

• Core Bridge: Initial PyO3 implementation and project stubs.
• Topographic Engine: Migration of adaptive Z-scaling logic from Python to Rust.
• Parallel Processing: Multi-threaded batch processing for the Paikuli corpus.
• OCR Pre-processing: Surface normal re-mapping for synthetic raking-light augmentation.

📜 Project Identity & IP

HERZFELD is developed by Andrea Marruzzo, Research Fellow at Sapienza University of Rome, as part of the study of the Middle Persian and Parthian Paikuli Inscription.

The choice of Rust for the core engine serves a dual purpose: providing the performance required for modern ML pipelines while protecting the intellectual property and algorithmic integrity of the research within a competitive academic environment.

⚖️ Citation

If you use the HERZFELD suite or the Rust core in your research, please cite it as:

APA: Marruzzo, A. (2026). HERZFELD (Suite): High‑fidelity Epigraphic Rendering (Version 0.1.0) [Software]. PyPI. https://pypi.org/project/herzfeld/

BibTeX:

@software{marruzzo_herzfeld_suite_2026,
  author = {Marruzzo, Andrea},
  title = {HERZFELD (Suite): High‑fidelity Epigraphic Rendering},
  version = {0.1.0},
  year = {2026},
  month = {1},
  publisher = {PyPI},
  url = {https://crates.io/crates/herzfeld},
  note = {Reserved namespace for HERZFELD rendering suite}
}