Cervix Locator + OOD Dataset
Dataset Description
⚠️ Warning: This dataset contains medical images of the cervix. Some images may be graphic or disturbing.
This dataset is designed for developing ML models to classify cervix types and to include out-of-distribution (OOD) images to test model robustness.
The original cervical images are sourced from the Intel & MobileODT Cervical Cancer Screening dataset, while the OOD images come from the Caltech-101 dataset.
Cervix Types
The cervix types in this dataset are all considered normal (not cancerous), but correct identification is critical for patient care and healthcare provider decision-making. Identifying transformation zones is challenging, so algorithm-aided detection improves screening efficiency and quality.
| Type | Description |
|---|---|
| Type 1 | Completely ectocervical; fully visible; small or large |
| Type 2 | Has endocervical component; fully visible; may have ectocervical component (small or large) |
| Type 3 | Has endocervical component; not fully visible; may have ectocervical component (small or large) |
Out-of-Distribution (OOD) Images
OOD images are sampled from Caltech-101, which contains:
- 101 object categories (e.g., “helicopter”, “elephant”, “chair”)
- A background category with images not belonging to the 101 categories
- Each category contains ~40–800 images (most ~50)
- Resolution ~300×200 pixels
These images are included to test model robustness to non-cervix images.
Dataset Structure
- Total images: 2,081
- Classes:
type1,type2,type3,ood - Shuffled and labeled in a DataFrame with columns:
image: Path to image filelabel: Class label
Class Distribution
| Class | Count |
|---|---|
| type1 | 250 |
| type2 | 781 |
| type3 | 450 |
| ood | 600 |
- Downloads last month
- 4