LRP for ResNet returns weird heatmap

Captum
1

Hey,

I’m trying to use your LRP implementation on a pretrained model resnet34/32 (ImagetNet and CIFAR10/100 respectively).

I needed to convert the ReLU layers to inplace=False in order to make it work and then when I execute this - it works but the output is weird and inconsistent…

these are the images for example:

image
image805×390 40.8 KB

and these are the explanations (heat maps)

image
image806×395 45.6 KB

which doesnt make sense since the LRP is always readable and coherent with the images themselves…

thanks!

2

I have excactly the same problem, with the LRP ResNET. Also I’ve tried different rules. Did you found any solution?