Colonclip: Colonoscopy Image-Text Pairs for Medical CLIP Training

Colonclip is a multimodal training dataset for medical computer vision models, likely containing colonoscopy images paired with text. The dataset was uploaded by author ZoeTAN to Hugging Face and was last updated on June 3, 2026. It includes LMDB archives for training, label embeddings, and a separate testing set.

Use Cases

• Training a medical CLIP model based on the described image-text pairs.
• Fine-tuning computer vision models for colonoscopy image classification using the provided label embeddings.
• Benchmarking multimodal retrieval systems on medical data using the separate test set.

Strengths

• Includes a dedicated testing set (test_ex_adapt_collect.zip) for evaluation.
• Provides label embeddings (label_en_dict.npy, label_en_graph.npz) which can facilitate structured learning.
• Offers data in multiple formats (LMDB, JSON, TSV) for different processing needs.

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.

Provenance

Source

ZoeTAN on Hugging Face

Freshness

Last updated 2026-06-03 11:00:54; freshness should be verified.