IDRiD is a clinically curated dataset of fundus images preserving native resolution and raw intensity. The dataset includes images graded for diabetic retinopathy severity levels from DR 0 to DR 4. The collection is hosted on Kaggle.
Use Cases
- Train diabetic retinopathy classification models based on clinically graded fundus images.
- Benchmark image processing algorithms based on native resolution and raw intensity data.
- Develop computer-aided diagnosis tools based on the severity levels DR 0-4.
Strengths
- Images are clinically curated, indicating a professional medical review process.
- Data preserves native resolution and raw intensity, which is beneficial for detailed analysis.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.