How can i use it with fully convolutional network (U-net) ??

[rameezrehman83]

Which changes should i make to use it with unet which has a sigmoid layer at the end and no softmax and fully connected layer?

[totti0223]

I cannot understand what you are trying to do. gradcam (and gradcampp) is for visualizing the intermediate map behind the classification problem.

[rameezrehman83]

I want to visualize the gradcam to visualize the last layer of U-net. In VGG16 there are fully connected and softmax layer to output the probability of a class, but in U-net (which is fully convolutional) there is no fully connected layer, just sigmoid layer after convolution layer outputting the segmented image. so i want to visualize the gradcam of layers, i want guidance about how can i use it with U-net model?? I am using this model, https://github.com/wuyang0329/unet my input images are grayscale, with binary labels.

[totti0223]

u-net is a segmentation (pixel-wise classification) network, not like a "classification" like in VGG16. it is not inpossible to apply gradcam or gradcampp against the single pixel of the output, but that is a nonsense. i prefer you to go through the blogs somewhere and get precise understanding of what you are trying to do in the first place. maybe simple visualisation of intermediate output is what you want to do.