Struct GradCAM 

pub struct GradCAM { /* private fields */ }

GradCAM (Gradient-weighted Class Activation Mapping)

GradCAM generates visual explanations for CNN decisions by using gradients flowing into the final convolutional layer to produce a coarse localization map highlighting important regions in the image.

Reference: Selvaraju et al., “Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization”, ICCV 2017

Implementations

impl GradCAM

pub fn new(target_layer_name: String, device: Arc<dyn Device>) -> Self

Create a new GradCAM explainer

Arguments

• target_layer_name - Name of the convolutional layer to use for visualization (typically the last conv layer before classification)
• device - Device to perform computations on

pub fn generate_heatmap( &self, model: &dyn Module, input: &Tensor, target_class: usize, ) -> Result<Tensor>

Generate GradCAM heatmap for a specific class

Arguments

• model - The model to explain
• input - Input image tensor [C, H, W] or [N, C, H, W]
• target_class - Class index to generate explanation for

Returns

Heatmap tensor of the same spatial dimensions as the input

pub fn generate_gradcam_plus_plus( &self, model: &dyn Module, input: &Tensor, target_class: usize, ) -> Result<Tensor>

Generate GradCAM++ heatmap (improved version of GradCAM)

GradCAM++ provides better localization and works better for multiple instances of the same class in an image.

pub fn overlay_heatmap( &self, image: &Tensor, heatmap: &Tensor, alpha: f32, ) -> Result<Tensor>

Overlay heatmap on original image

Arguments

• image - Original image tensor [C, H, W]
• heatmap - Heatmap tensor [H, W]
• alpha - Blending factor (0.0 = only image, 1.0 = only heatmap)

Returns

Blended visualization [C, H, W]