Camelyon16 Uni: Patch Embeddings for Histopathology Images

Patch embeddings for the CAMELYON16 dataset generated using the UNI foundation model. The embeddings are derived from 128x128 micrometer tissue patches, with segmentation and patching performed using a modified version of the CLAM toolkit. The dataset was authored by kaczmarj and last updated on December 10, 2025.

Use Cases

• Training or fine-tuning machine learning models for cancer detection based on patch-level image embeddings.
• Developing self-supervised or transfer learning approaches for histopathology using pre-computed foundation model features.
• Benchmarking the performance of different embedding methods on the established CAMELYON16 challenge tasks.
• Analyzing tissue morphology and patterns at a constant physical scale across whole-slide images.

Strengths

• Embeddings are generated using a modern foundation model (UNI), which likely provides high-quality feature representations.
• Patches are extracted at a consistent physical size (128x128 micrometers), enabling standardized spatial analysis.
• Builds upon the well-known CAMELYON16 benchmark dataset for metastatic tissue detection.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Based on the CAMELYON16 dataset, processed by author kaczmarj.

Collection Method

Tissue segmentation and patching performed with a modified version of the CLAM toolkit, embeddings created with the UNI foundation model.

Freshness

Last updated 2025-12-10 20:39:38; freshness should be verified.