Skip to main content
. 2022 Apr 28;13:864486. doi: 10.3389/fpls.2022.864486

FIGURE 12.

FIGURE 12

Visualization results of the three schemes. (A) Grad-Cam method for visualization of the feature maps output by the Resnet50 convolutional layer. (B) Grad-Cam method for visualization of the feature maps output from the max-pooling layer of the standard model. (C) Grad-Cam method for visualization of the tokens of the standard model. (D) Directly map the attention of the classification token on feature tokens to the original image.