OLDA-Bench Dataset
Overview
OLDA-Bench is a benchmark dataset for oral leukoplakia progression analysis. The dataset contains oral photographic images with lesion annotations, covering oral leukoplakia and oral leukoplakia-associated cancer lesions.
The dataset is designed to support:
- Oral lesion detection
- Instance segmentation
- Progression recognition
- Clinical validation studies
Dataset Structure
OLDA-Bench-Dataset
│
├── Public_all_train
├── PPT_valid
└── Clinical_test
Public_all_train
Training set collected from publicly available oral lesion resources.
PPT_valid
Validation set used for model development and hyperparameter tuning.
Clinical_test
Independent external clinical test set collected from real-world clinical practice.
Categories
The dataset contains two categories:
| Category | Description |
|---|---|
| Leukoplakia | Oral leukoplakia lesions |
| Leukoplakia-Cancer | Oral leukoplakia lesions with malignant transformation |
Annotation Format
Images are provided in JPEG format:
xxx.jpeg
Annotations are provided in Labelme JSON format:
xxx.json
Each JSON file contains polygon-level lesion annotations corresponding to its image.
Access Policy
The dataset is released under gated access.
Researchers may request access through the Hugging Face dataset page.
Access requests are reviewed manually and granted for academic and non-commercial research purposes only.
Related Resources
Code Repository
https://github.com/GuanyuChen8/OLDA-Bench
Model Repository
https://huggingface.co/gy-chen/OLDA-Bench
- Downloads last month
- 4