How can I access layer parametr as eg. convolution_param.num_output in python API

[Jan Pohanka]

Hi,

I cannot found how to access layer parameters from python

net = caffe.Net(model_filename, weight_filename, caffe.TEST)
net.layers[0]. ???

Is there such API?

thanks

Jan

[Mark]

Here's one option if you are reading in a protobuf file and want to access the parameters.

import caffe
from google.protobuf import text_format

net_config = caffe.proto.caffe_pb2.NetParameter()

train_file = "/path/to/file/train.prototxt"

file = open(train_file, "r")
text_format.Merge(str(file.read()), net_config)
file.close()

# At this point you can access the layer information from the net_config object
# For example get all the layers in the network with:
layer_list = net_config.layer

# View all the layers with protobuf formatting:
for layer in layer_list:
    print(layer)

# Assuming layer 5 is a convolution parameter you can access the num_output parameter with:
conv_num_out = layer_list[5].convolution_param.num_output

You can access any parameter of the network this way.

    -Mark

[Jan Pohanka]

Hi Mark,

thanks for suggestion. This might be a solution even if it not seems very elegant to me..

Dne čtvrtek 11. května 2017 19:00:49 UTC+2 Mark napsal(a):