README.md

---
license: mit
size_categories:
- 1K<n<10K
pretty_name: STEM2Mat
tags:
- chemistry
task_categories:
- image-to-3d
---

# AutoMat Benchmark: STEM Image to Crystal Structure

The **AutoMat Benchmark** is a multimodal dataset designed to evaluate deep‑learning systems for iDPC-STEM‑based crystal‑structure reconstruction and property prediction.

Code: https://github.com/yyt-2378/AutoMat

---

## 📁 Dataset Structure

The dataset is organized into three tiers of increasing difficulty:

```text
benchmark/
├── tier1/
│   ├── img/          # STEM images (e.g., PNG, TIFF)
│   ├── label/        # Atomic position labels (e.g., TXT, JSON)
│   └── cif_file/     # Reconstructed or ground‑truth CIF files
├── tier2/
│   └── ...           # Same sub‑folders as tier1
├── tier3/
│   └── ...
└── property.csv      # Material properties for all samples
````

---

## 🔬 Tier Descriptions

| Tier   | Characteristics                                                |
|--------|----------------------------------------------------------------|
| Tier 1 | Simulated low-noise STEM images, light elements, low complexity |
| Tier 2 | Moderate noise or multiple elements, more realistic patterns   |
| Tier 3 | Low dose, multi-elements, complex symmetry                     |

Each sample in the dataset includes:
- A STEM image (`img/`)
- Labeled atomic coordinates (`label/`)
- A reconstructed or reference CIF file (`cif_file/`)
- Associated material properties in `property.csv`

---

## 📊 Tasks Supported

- STEM-to-structure inference
- CIF generation and comparison
- Atomic position prediction
- Property prediction (formation energy, energy_per_atom, bandgap, etc.)

---

## 🔗 Files Description

| File / Folder      | Description                                                |
|--------------------|------------------------------------------------------------|
| `img/`             | STEM input images (grayscale microscopy)                   |
| `label/`           | Atomic position data (format: .png per sample)     |
| `cif_file/`        | .cif crystal structure files for each sample               |
| `property.csv`     | Global material properties table with `material_id` match  |

---

## 📄 License

This dataset is released under the **MIT License**. You are free to use, modify, and distribute with attribution.

---

## ✉️ Citation

If you use this benchmark, please cite:

```bibtex
@misc{yang2025automatenablingautomatedcrystal,
      title={AutoMat: Enabling Automated Crystal Structure Reconstruction from Microscopy via Agentic Tool Use}, 
      author={Yaotian Yang and Yiwen Tang and Yizhe Chen and Xiao Chen and Jiangjie Qiu and Hao Xiong and Haoyu Yin and Zhiyao Luo and Yifei Zhang and Sijia Tao and Wentao Li and Qinghua Zhang and Yuqiang Li and Wanli Ouyang and Bin Zhao and Xiaonan Wang and Fei Wei},
      year={2025},
      eprint={2505.12650},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2505.12650}, 
}
```

---

## 🙋 Contact

For questions or collaborations, please contact: `yangyt22@gmail.com`