Cannot import custom layers in jupyter notebook
adhara123007 opened this issue · comments
Thanks for being active here in resolving queries.
This is more of a question since I am a little inexperienced with CAFFE.
I wanted to understand the functions of your custom layers like : resample_layer and data_augmentation layer (not present in the documentation).
So, I created a protxt file with these layers
name: "CaffeNet"
layer {
name: "data"
type: "Input"
top: "data"
input_param { shape: { dim: 1 dim: 1 dim: 227 dim: 227 } }
}
layer {
name: "Eltwise1"
type: "Eltwise"
bottom: "data"
top: "data_sum"
eltwise_param {
operation: SUM
coeff: 0.00392156862745098
}
}
layer {
name: "Resample1"
type: "Resample"
bottom: "data_sum"
top: "data_resample"
resample_param {
width: 1024
height: 1792
type: LINEAR
antialias: true
}
}
When I try to use this network in jupyter notebook it fails. However, the layers defined in the documentation (Eltwise and Convolution) work well with the jupyter notebook.
import caffe
model_def = caffe_root + 'examples/augment.prototxt'
net = caffe.Net (model_def, caffe.TEST)
net.blobs['data'].data[...] = im_input
output = net.forward()
As soon as I run this the kernel becomes dead.
What I am missing here ?
Your example works in my Python terminal (I don't have a Jupyter setup), but I had to enable GPU mode (EDIT: because our puny layer implementations do not have CPU code..):
caffe.set_device(1)
caffe.set_mode_gpu()
Is it possible that you have multiple Caffe installations, and that the wrong Python module / .so file is loaded?
It works with my Python terminal . I will check again and close this issue.
Works