SLAKE
SLAKE is a medical dataset designed for Visual Question Answering (VQA) applied to medical imaging, allowing AI models to learn how to answer questions about clinical images.
14,028 annotated images, JSON files, and Parquet, approximately 4.2 MB
CC-BY 4.0
Description
SLAKE is a semantically labeled data set designed for Visual Question Answering in medical imaging. It contains over 14,000 sample images annotated with specific questions and answers from a variety of medical modalities.
What is this dataset for?
- Train models capable of answering questions about medical images
- Improving diagnostic support systems through the analysis of clinical images
- Evaluate the visual understanding of AI models in a specialized medical context
Can it be enriched or improved?
Yes, this dataset can be supplemented with other imaging modalities or enriched with additional annotations such as segmentations, more accurate clinical labels, or anonymized patient data. Integrating temporal metadata can also strengthen use cases.
🔎 In summary
🧠 Recommended for
- Medical AI researchers
- Diagnostic support system developers
- VQA teams
🔧 Compatible tools
- PyTorch
- TensorFlow
- MONAI
- Detectron2
- MedVQA frameworks
💡 Tip
To improve performance, combine SLAKE with multi-modal annotated datasets and learning transfer techniques.
Frequently Asked Questions
Is SLAKE suitable for non-medical use?
No, this dataset is specialized in medical imaging and is not suitable for general non-health cases.
Can SLAKE be used to train image segmentation models?
Not directly, as the dataset targets the VQA, but additional annotations may allow for segmentation.
Is it a large dataset?
Yes, with over 14,000 examples, it offers enough volume for meaningful training.
