SCUT-CAB: A New Benchmark Dataset of Ancient Chinese Books with Complex Layouts for Document Layout Analysis

The SCUT-CAB dataset is a benchmark dataset for document layout analysis in Chinese ancient books, published in ICFHR 2022 by the Deep Learning and Visual Computing Lab of South China University of Technology (SCUT DLVC Lab).

---

⚠️ Access Request & Application Instructions

The SCUT-CAB dataset can only be used for non-commercial research purposes. The archive file is publicly accessible but encrypted with a password. To request the decompression password, please follow these steps:

Step 1: Download and complete the agreement document:

• 📄 Application Form for Using SCUT-CAB (Direct Download)

Have this document signed and stamped by your institution. Please also prepare 1–2 recent publications (within the last 6 years) as evidence that you or your team conduct research in OCR, handwriting analysis and recognition, document image processing, or visual information extraction.

Step 2: Submit your application online:

🔗 SCUT DLVC Lab Dataset Access Portal → Apply for SCUT-CAB

Upload both signed documents through the portal and fill out the "Recent Publications" block. Your application will be reviewed manually and you will be notified by email once a decision has been made (typically within 1–5 business days).

Step 3: Decompress the dataset:

After approval, you will receive the decompression password via email.

⚠️ All users must comply with the use conditions at all times; failure to do so will result in revocation of access.

---

Dataset Download

Platform	Download Link	Format / Size	Password Required
Hugging Face	🤗 hiuyi/SCUT-CAB Repository	SCUT_CAB.rar	Yes (encrypted archive)
Baidu Cloud	Download via BaiduNetdisk	6.84 GB (Extract Code: dlvc)	Yes (decompression password)

How to Download from Hugging Face

You can download SCUT_CAB.rar using any of the following methods:

Method 1: Direct Web Download

• Go directly to the repository file list: hiuyi/SCUT-CAB / Files and versions and click download on SCUT_CAB.rar.

Method 2: Using Python (huggingface_hub)

from huggingface_hub import hf_hub_download

hf_hub_download(
    repo_id="hiuyi/SCUT-CAB",
    filename="SCUT_CAB.rar",
    repo_type="dataset",
    local_dir="./"
)

Method 3: Using Hugging Face CLI

huggingface-cli download --repo-type dataset hiuyi/SCUT-CAB SCUT_CAB.rar --local-dir ./

---

Dataset Overview

The SCUT-CAB dataset contains 4,000 manually annotated ancient book images with a total of 31,925 layout element annotations, covering various binding formats, fonts, and preservation qualities.

Subsets:

• SCUT-CAB-Logical (Logical Layout Analysis): 27 fine-grained semantic categories: {EOV, author, bibliography, book number, caption, centerfold strip, chapter title, collation table, colophon, compiler, ear note, endnote, engraver, figure, foliation, header, interlinear note, marginal annotation, page box, part, section title, sub section title, subtitle, sutra number, text, title, volume number}.
• SCUT-CAB-Physical (Physical Layout Analysis): 4 primary physical categories: {centerfold strip, figure, page box, text}.
• Reading Order: Includes reading-order annotations (traditional top-to-bottom, right-to-left order for body text).

Data Sources & Layouts:

• Sources: Buddhist scriptures (Tripitaka), Reproductions of the Chinese Rare Editions Series, and Local chronicles.
• Bindings: Warp-fold binding, Wrapped-back binding, Photocopies, Butterfly binding, etc.
• Degradation Diversity: Damaged pages, ink fading, and bleed-through / back-through effects.

---

Directory Format

Once SCUT_CAB.rar is decompressed, the dataset is organized in the following format:

├── SCUT_CAB
    ├── images
    │   ├── train
    │   │   ├── xxx.jpg
    │   │   └── ...
    │   └── test
    │       ├── xxx.jpg
    │       └── ...
    ├── SCUT_CAB_logical
    │   ├── labels_logical.txt
    │   ├── annotations
    │   │   ├── instances_train.json
    │   │   └── instances_test.json
    │   └── json
    │       ├── xxx.json
    │       └── ...
    └── SCUT_CAB_physical
        ├── labels_physical.txt
        ├── annotations
        │   ├── instances_train.json
        │   └── instances_test.json
        └── json
            ├── xxx.json
            └── ...

---

License

The SCUT-CAB dataset should be used and distributed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) License for non-commercial research purposes.

---

Citation and Contact

Please cite our paper if you use this dataset in your research:

@article{cheng2022scut,
  title={SCUT-CAB: A New Benchmark Dataset of Ancient Chinese Books with Complex Layouts for Document Layout Analysis},
  author={Hiuyi Cheng, Cheng Jian, Sihang Wu, Lianwen Jin},
  booktitle={International Conference on Frontiers of Handwriting Recognition (ICFHR)},
  year={2022}
}

• Modern Document Layout Analysis (Related Work): M6Doc Dataset Release | Hugging Face hiuyi/M6Doc
• Contact: For inquiries, please contact Prof. Lianwen Jin at eelwjin@scut.edu.cn or lianwen.jin@gmail.com.