Image Segmentation Model Scores for 200 Test Images

200 medical images were used to generate Intersection over Union (IoU), Dice, and peak-activation scores for evaluating a segmentation model. Md Shihab Reza published this small, curated subset of performance metrics in April 2026. Aggregate statistics, including mean and 95% confidence intervals, are provided for the full test set.

Use Cases

• Benchmark segmentation model performance by comparing IoU and Dice scores across different architectures.
• Analyze the correlation between peak-activation scores and segmentation accuracy (IoU/Dice) for specific image regions.
• Validate model consistency by examining the 95% confidence intervals reported for aggregate metrics on the full 200-image set.
• Use the provided mean scores as a baseline for new model evaluations on similar medical imaging tasks.

Strengths

• Dataset provides three standard segmentation evaluation metrics (IoU, Dice, peak-activation) per image.
• Aggregate statistics (mean and 95% CI) are computed over a defined test set of 200 images.

Limitations

• The shared file is only 5.5 KB and contains a curated subset, not the full 200-image data or raw predictions.
• No raw image data or model outputs are included, only derived numerical scores.
• Unknown sample size (rows) within the provided subset file.

Provenance

Source

Md Shihab Reza via figshare.

Collection Method

Metrics computed from a model's performance on a test set of medical images.

Freshness

Last updated April 2026.