search

raghakot / keras-vis

Neural network visualization toolkit for keras
https://raghakot.github.io/keras-vis
MIT License
2.98k stars 660 forks source link

saliency + guided backprop + all filters + linear activation + VGG16 gives zero gradients #200

Open eyaler opened 5 years ago

eyaler commented 5 years ago

not sure why but the combination of guided backprop + filter_indices=None (all filters) + linear final activation gives me zero gradients for VGG16. you can see this in the demo notebook.

eyaler commented 5 years ago

i now also see this sometimes with softmax