Attention Module Comparison Metrics for Inception-ResNetV2

A 9.6 KB Excel table contrasts the Squeeze-and-Excitation (SE) attention module with six alternatives (ACmix, CA, SA, KNNA, ECA, CBAM) within an Inception-ResNetV2 model. The table, authored by Chao Zhang and last updated in May 2026, contains pairwise DeLong P values, calibration metrics, and net-benefit indices. The SE module achieved the lowest Brier score and log-loss and the highest decision-curve metrics.

Use Cases

• Benchmarking attention module performance based on statistical significance tests.
• Evaluating model calibration for clinical utility based on Brier score and log-loss metrics.
• Comparing decision-curve analysis outcomes across different attention mechanisms.

Strengths

• Directly compares six specific attention modules (ACmix, CA, SA, KNNA, ECA, CBAM) against a Squeeze-and-Excitation baseline.
• Includes multiple evaluation metrics: DeLong P values, Brier score, log-loss, and net-benefit indices.
• Released under a permissive CC-BY-4.0 license for open use.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The description is limited to summary results; the raw data underlying the metrics is not included.

Provenance

Source

Chao Zhang via figshare

Collection Method

Likely generated from model evaluation experiments.

Freshness

Last updated 2026-05-14 17:44:30