3D CNN grad-cam: failing in scale_cam_image function

[neeleshbisht99]

I'm trying to use pytorch_grad_cam for a 3D CNN model. But it is failing maybe because of this code:

if len(img.shape) > 3:
                img = zoom(np.float32(img), [
                           (t_s / i_s) for i_s, t_s in zip(img.shape, target_size[::-1])])
            else:
                img = cv2.resize(np.float32(img), target_size)

Shoudln't the condition be len(img.shape) >= 3 for 3d datasets ?

All the code for the model and the error trace (in the last section) can be seen here : https://nbviewer.org/gist/neeleshbisht99/8bdc320fc49c69f3b90af06c9ceab2fa

[neeleshbisht99]

Also, it would be very helpful for many people if we could include a working example for 3D CNNs. It's evident from the number of issue that have opened in the past, that lots of people are facing issues while using the package for 3D CNNs.

Maybe we can add the above gist as an example for 3D CNN, in the repo itself, once we get that gist in working condition.

what say @jacobgil ?