Backward Pass through the net gives an array of zeros

[Mohit Jain]

Hi,
I'm facing an issue while computing the backward pass through a net. While using the feature_extraction ipython notebook sample given by caffe, I'm setting the 'force_backward: true' variable in my deploy.prototxt file, however, the backward pass still gives an array full of 0's. Can someone help me out as to what am I doing wrong?

REFERENCE : https://github.com/BVLC/caffe/issues/583

CODE:

caffe.set_mode_cpu()
net = caffe.Net(caffe_root + 'models/bvlc_reference_caffenet/forceback_deploy.prototxt',
caffe_root + 'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel',
caffe.TEST)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2,0,1))
transformer.set_mean('data', np.load(caffe_root + 'python/caffe/imagenet/ilsvrc_2012_mean.npy').mean(1).mean(1)) # mean pixel
transformer.set_raw_scale('data', 255) 
transformer.set_channel_swap('data', (2,1,0))

net.blobs['data'].reshape(1,3,227,227)
net.blobs['data'].data[...] = transformer.preprocess('data', caffe.io.load_image(caffe_root + 'examples/images/fish-bike.jpg'))
out = net.forward()
print("Predicted class is #{}.".format(out['prob'].argmax()))
back = net.backward()
j = back['data'].copy()
print j
^^^^^^^^^^^^^^^^^^^^^^^ THIS GIVES ME AN ARRAY OF ZEROS ^^^^^^^^^^^^^^^^^

forceback_deploy.prototxt

name: "CaffeNet"
input: "data"
input_dim: 10
input_dim: 3
input_dim: 227
input_dim: 227
force_backward: true
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output: 96
kernel_size: 11
stride: 4
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "conv1"
top: "conv1"
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "norm1"
type: "LRN"
bottom: "pool1"
top: "norm1"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "norm1"
top: "conv2"
convolution_param {
num_output: 256
pad: 2
kernel_size: 5
group: 2
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "norm2"
type: "LRN"
bottom: "pool2"
top: "norm2"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "norm2"
top: "conv3"
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "conv4"
type: "Convolution"
bottom: "conv3"
top: "conv4"
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
group: 2
}
}
layer {
name: "relu4"
type: "ReLU"
bottom: "conv4"
top: "conv4"
}
layer {
name: "conv5"
type: "Convolution"
bottom: "conv4"
top: "conv5"
convolution_param {
num_output: 256
pad: 1
kernel_size: 3
group: 2
}
}
layer {
name: "relu5"
type: "ReLU"
bottom: "conv5"
top: "conv5"
}
layer {
name: "pool5"
type: "Pooling"
bottom: "conv5"
top: "pool5"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "fc6"
type: "InnerProduct"
bottom: "pool5"
top: "fc6"
inner_product_param {
num_output: 4096
}
}
layer {
name: "relu6"
type: "ReLU"
bottom: "fc6"
top: "fc6"
}
layer {
name: "drop6"
type: "Dropout"
bottom: "fc6"
top: "fc6"
dropout_param {
dropout_ratio: 0.5
}
}
layer {
name: "fc7"
type: "InnerProduct"
bottom: "fc6"
top: "fc7"
inner_product_param {
num_output: 4096
}
}
layer {
name: "relu7"
type: "ReLU"
bottom: "fc7"
top: "fc7"
}
layer {
name: "drop7"
type: "Dropout"
bottom: "fc7"
top: "fc7"
dropout_param {
dropout_ratio: 0.5
}
}
layer {
name: "fc8"
type: "InnerProduct"
bottom: "fc7"
top: "fc8"
inner_product_param {
num_output: 1000
}
}
layer {
name: "prob"
type: "Softmax"
bottom: "fc8"
top: "prob"
}

[Mohit Jain]

One needs to set the diff of the last layer to 1 at the index of our target class. So, if you are doing imagenet classification with the tabby cat, after doing the forward pass, do something like this : 

net.blobs['prob'].diff[0][281]=1   #281 is tabby cat.

and then extract the weights via backpass through the network.

PS : Don't forget to put a "force_backward: true" in your deploy.prototxt file.

Regards,
Mohit

[Evan Shelhamer]

This is the right approach -- I'm commenting to confirm since there were a few of these questions lately. Thanks for following up with your own answer.

Evan Shelhamer

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users...@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/6989158f-2360-427e-824b-9c6162263e8b%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.