Attention Supervision

Attention supervision mechanisms for interpretable medical imaging models.

Core Components

The attention supervision module provides mechanisms to guide model attention using supervision signals, improving both interpretability and performance.

Supervision Strategies

Various supervision strategies are available including region-based, point-based, and scribble-based supervision for different annotation types.

Usage Example

from med_core.attention_supervision import AttentionSupervision

# Create attention supervision
supervision = AttentionSupervision(
    supervision_type='region',
    loss_weight=0.1
)

# Apply supervision
attention_maps = model.get_attention_maps()
supervision_loss = supervision(attention_maps, region_masks)