MEDISEG: Medication Images with Instance Segmentation Masks

MEDISEG contains two subsets for pill recognition: a controlled 3-pill set and a diverse 32-pill set with real-world challenges. The dataset includes 426.5 MB of high-resolution JPG images with COCO-format annotations for segmentation masks, bounding boxes, and class labels. William Chu published this dataset on figshare in May 2026.

Use Cases

• Train pill segmentation models based on instance segmentation masks and bounding boxes.
• Evaluate model performance under real-world conditions based on varied lighting, occlusions, and cluttered scenes.
• Benchmark few-shot learning algorithms based on the MEDISEG-Deploy subset with fixed support and query partitions.
• Develop medication verification systems based on the provided metadata including drug registration numbers and active ingredients.

Strengths

• Includes two distinct subsets: a controlled 3-pill set and a more diverse 32-pill set.
• Provides COCO-format annotations for segmentation masks, bounding boxes, and class labels.
• MEDISEG-Deploy subset is designed for few-shot evaluation under 1-shot, 5-shot, and 10-shot settings.
• Contains supplementary metadata.csv with drug information like brand names and active ingredients.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

figshare

Freshness

Last updated 2026-05-12 12:26:31; freshness should be verified.