leftthomas / gradcam Goto Github PK

I tried the program on the resnet 34 architecture with 100 classes and it simply did not work.
Apparently, the matrix transpose operation could not be performed. The size mismatch was 512x1 to 512x100.

Hi, @leftthomas

As a result of classifying with Resnet, Accuarcy is over 99%. If you hit map the object area with gradCAM with that model file, it does not match exactly. Why?
it does not match exactly. Why?

It seems to be a problem of GradCAM rather than Resnet classification learning. The objects to be hit-mapped are not as local or blob like dogs or cats, but close to a long straight line. In this case, GradCAM seems to miss the object area. Have you experienced this?

For a well-trainedd Resnet34 model, how do you optimize GradCAM?

Thanks, in advance.

from @bemoregt.

            classes = F.sigmoid(feature) #feature shape=[1,1], classes shape=[1,1]
            one_hot, _ = classes.max(dim=-1) # one_hot shape=[1]
            self.model.zero_grad()
            one_hot.backward()

This one_hot is not the one hot label, I don't know it is correct or not,

In my experiment, there is a class without any activation hot map. Did you know why? Thanks

Hi, @mingloo @jacobgil @jdecid @ChaiKnight @flyingpot

I have fine-tuned VGG19_bn on my custom image set.
Please guide how to modify the code to use my model.

Thank you!
Best.
@bemoregt.