Libre-YOLO/paper-parts

Paper Parts

This dataset is part of the Roboflow 100 benchmark, a diverse collection of 100 object detection datasets spanning 7 imagery domains.

Dataset Statistics

Split	Images
Train	8,472
Validation	2,359
Test	1,209
Total	12,040

Classes (46)

• abstract heading
• abstract text
• algorithm
• author
• chapter
• claim number
• claim
• comittee
• corollary number
• corollary
• date
• definition number
• definition
• degree
• equation number
• equation
• example caption
• example
• figure caption
• figure title
• figure
• footnote
• lemma number
• lemma
• list of content heading
• list of content text
• page number
• paragraph
• proposition number
• proposition
• reference text
• reference title
• scheme caption
• scheme
• section
• subsection
• subsubsection
• table caption
• table of contents text
• table of contents title
• table title
• table
• theorem number
• theorem
• title
• university

Usage

With LibreYOLO

from libreyolo import LIBREYOLO

# Load a model
model = LIBREYOLO(model_path="libreyoloXnano.pt")

# Train on this dataset
model.train(data='path/to/data.yaml', epochs=100)

Download from HuggingFace

from huggingface_hub import snapshot_download

# Download the dataset
snapshot_download(
    repo_id="Libre-YOLO/paper-parts",
    repo_type="dataset",
    local_dir="./paper-parts"
)

Directory Structure

paper-parts/
├── data.yaml           # Dataset configuration
├── README.md           # This file
├── train/
│   ├── images/         # Training images
│   └── labels/         # Training labels (YOLO format)
├── valid/
│   ├── images/         # Validation images
│   └── labels/         # Validation labels
└── test/
    ├── images/         # Test images (if available)
    └── labels/         # Test labels

Label Format

Labels are in YOLO format (one .txt file per image):

<class_id> <x_center> <y_center> <width> <height>

All coordinates are normalized to [0, 1].

Citation

If you use this dataset, please cite the Roboflow 100 benchmark:

@misc{rf100_2022,
    Author = {Floriana Ciaglia and Francesco Saverio Zuppichini and Paul Guerrie and Mark McQuade and Jacob Solawetz},
    Title = {Roboflow 100: A Rich, Multi-Domain Object Detection Benchmark},
    Year = {2022},
    Eprint = {arXiv:2211.13523},
}

License

This dataset is released under the CC-BY-4.0 license. Please check the original source for any additional terms.

Acknowledgments

• Original dataset from Roboflow Universe
• Part of the Roboflow 100 Benchmark
• Sponsored by Intel