Do not get exact Average drop in GradCAM & GradCAM++

[tfc-0]

Hi,

I tried the same experiment on the CAM methods for 50K images of ImageNet using torchvision's ResNet50 model.

Differently from previous works which conducted the evaluation on randomly selected images, we conduct experiments on the entire ImageNet validation set (ILSVRC2012) [25] consisting of 50000 images [mentioned in your paper]

However, my Average drop in confidence values is:

GradCAM: 14.177663100940071
GradCAM++: 14.740725754849704
SmoothGradCAM++: 16.673648117549764
ScoreCAM: 8.69675494321261

I used the same torch-cam package that you used. My GradCAM and GradCAM++ 's values are almost the same, whereas your values have a lot of differences. Your values:

Can you tell me which mistakes I have done?