Skip to content

GitLab

"tutorials/vscode:/vscode.git/clone" did not exist on "bea07b41b3429f841ed1599ed25bd40ea86ac9da"
Commit cbda2b9e authored by Davis King's avatar Davis King
Browse files

Changed caffe converter to require the user to specify the input tensor size 

when the converter is run.
parent 523e020e
Hide whitespace changes
Inline Side-by-side
Showing with 47 additions and 24 deletions
tools/convert_dlib_nets_to_caffe/main.cpp
View file @ cbda2b9e
......@@ -105,7 +105,11 @@ void print_as_np_array(std::ostream& out, const matrix_exp<EXP>& m)
// ----------------------------------------------------------------------------------------
void convert_dlib_xml_to_caffe_python_code(
const string& xml_filename
const string& xml_filename,
const long N,
const long K,
const long NR,
const long NC
)
{
const string out_filename = left_substr(xml_filename,".") + "_dlib_to_caffe_model.py";
......@@ -124,26 +128,34 @@ void convert_dlib_xml_to_caffe_python_code(
// dlib nets don't commit to a batch size, so just use 1 as the default
fout << "\n# Input tensor dimensions" << endl;
fout << "batch_size = 1;" << endl;
fout << "input_batch_size = " << N << ";" << endl;
if (layers.back().detail_name == "input_rgb_image")
{
cout << "WARNING: The source dlib network didn't commit to a specific input tensor size, we are using a default size of 28x28x1 which is appropriate for MNIST input. But if you are using different inputs you will need to edit the auto-generated python script to tell it your input size." << endl;
fout << "input_nr = 28; #WARNING, the source dlib network didn't commit to a specific input size, so we put 28 here as a default. It might not be the right value." << endl;
fout << "input_nc = 28; #WARNING, the source dlib network didn't commit to a specific input size, so we put 28 here as a default. It might not be the right value." << endl;
fout << "input_k = 3;" << endl;
fout << "input_num_channels = 3;" << endl;
fout << "input_num_rows = "<<NR<<";" << endl;
fout << "input_num_cols = "<<NC<<";" << endl;
if (K != 3)
throw dlib::error("The dlib model requires input tensors with NUM_CHANNELS==3, but the dtoc command line specified NUM_CHANNELS=="+to_string(K));
}
else if (layers.back().detail_name == "input_rgb_image_sized")
{
fout << "input_nr = " << layers.back().attribute("nr") << ";" << endl;
fout << "input_nc = " << layers.back().attribute("nc") << ";" << endl;
fout << "input_k = 3;" << endl;
fout << "input_num_channels = 3;" << endl;
fout << "input_num_rows = " << layers.back().attribute("nr") << ";" << endl;
fout << "input_num_cols = " << layers.back().attribute("nc") << ";" << endl;
if (NR != layers.back().attribute("nr"))
throw dlib::error("The dlib model requires input tensors with NUM_ROWS=="+to_string(layers.back().attribute("nr"))+", but the dtoc command line specified NUM_ROWS=="+to_string(NR));
if (NC != layers.back().attribute("nc"))
throw dlib::error("The dlib model requires input tensors with NUM_COLUMNS=="+to_string(layers.back().attribute("nc"))+", but the dtoc command line specified NUM_COLUMNS=="+to_string(NC));
if (K != 3)
throw dlib::error("The dlib model requires input tensors with NUM_CHANNELS==3, but the dtoc command line specified NUM_CHANNELS=="+to_string(K));
}
else if (layers.back().detail_name == "input")
{
cout << "WARNING: The source dlib network didn't commit to a specific input tensor size, we are using a default size of 28x28x1 which is appropriate for MNIST input. But if you are using different inputs you will need to edit the auto-generated python script to tell it your input size." << endl;
fout << "input_nr = 28; #WARNING, the source dlib network didn't commit to a specific input size, so we put 28 here as a default. It might not be the right value." << endl;
fout << "input_nc = 28; #WARNING, the source dlib network didn't commit to a specific input size, so we put 28 here as a default. It might not be the right value." << endl;
fout << "input_k = 1;" << endl;
fout << "input_num_channels = 1;" << endl;
fout << "input_num_rows = "<<NR<<";" << endl;
fout << "input_num_cols = "<<NC<<";" << endl;
if (K != 1)
throw dlib::error("The dlib model requires input tensors with NUM_CHANNELS==1, but the dtoc command line specified NUM_CHANNELS=="+to_string(K));
}
else
{
......@@ -173,7 +185,7 @@ void convert_dlib_xml_to_caffe_python_code(
fout << " # For reference, the only \"documentation\" about caffe layer parameters seems to be this page:\n";
fout << " # https://github.com/BVLC/caffe/blob/master/src/caffe/proto/caffe.proto\n" << endl;
fout << " n = caffe.NetSpec(); " << endl;
fout << " n.data,n.label = L.MemoryData(batch_size=batch_size, channels=input_k, height=input_nr, width=input_nc, ntop=2)" << endl;
fout << " n.data,n.label = L.MemoryData(batch_size=input_batch_size, channels=input_num_channels, height=input_num_rows, width=input_num_cols, ntop=2)" << endl;
// iterate the layers starting with the input layer
for (auto i = layers.rbegin(); i != layers.rend(); ++i)
{
......@@ -389,15 +401,24 @@ void convert_dlib_xml_to_caffe_python_code(
int main(int argc, char** argv) try
{
if (argc == 1)
if (argc != 6)
{
cout << "Give this program an xml file generated by dlib::net_to_xml() and it will" << endl;
cout << "convert it into a python file that outputs a caffe model containing the dlib model." << endl;
cout << "To use this program, give it an xml file generated by dlib::net_to_xml() " << endl;
cout << "and then 4 numbers that indicate the input tensor size. It will convert " << endl;
cout << "the xml file into a python file that outputs a caffe model containing the dlib model." << endl;
cout << "For example, you might run this program like this: " << endl;
cout << " ./dtoc lenet.xml 1 1 28 28" << endl;
cout << "would convert the lenet.xml model into a caffe model with an input tensor of shape(1,1,28,28)" << endl;
cout << "where the shape values are (num samples in batch, num channels, num rows, num columns)." << endl;
return 0;
}
for (int i = 1; i < argc; ++i)
convert_dlib_xml_to_caffe_python_code(argv[i]);
const long N = sa = argv[2];
const long K = sa = argv[3];
const long NR = sa = argv[4];
const long NC = sa = argv[5];
convert_dlib_xml_to_caffe_python_code(argv[1], N, K, NR, NC);
return 0;
}
......
tools/convert_dlib_nets_to_caffe/running_a_dlib_model_with_caffe_example.py
View file @ cbda2b9e
......@@ -10,8 +10,10 @@ import numpy as np
# dlib lenet model. Then you need to convert that model into a "dlib to caffe
# model" python script. You can do this using the command line program
# included with dlib: tools/convert_dlib_nets_to_caffe. That program will
# output a lenet_dlib_to_caffe_model.py file. This line here imports that
# file.
# output a lenet_dlib_to_caffe_model.py file. You run that program like this:
# ./dtoc lenet.xml 1 1 28 28
# and it will create the lenet_dlib_to_caffe_model.py file, which we import
# with the next line:
import lenet_dlib_to_caffe_model as dlib_model
# lenet_dlib_to_caffe_model defines a function, save_as_caffe_model() that does
......@@ -54,12 +56,12 @@ data = np.array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,121,254,254,219,40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,121,254,207,18, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], dtype='float32');
data.shape = (dlib_model.batch_size, dlib_model.input_k, dlib_model.input_nr, dlib_model.input_nc);
data.shape = (dlib_model.input_batch_size, dlib_model.input_num_channels, dlib_model.input_num_rows, dlib_model.input_num_cols);
# labels isn't logically needed but there doesn't seem to be a way to use
# caffe's Net interface without providing a superfluous input array. So we do
# that here.
labels = np.ones((dlib_model.batch_size), dtype='float32')
labels = np.ones((dlib_model.input_batch_size), dtype='float32')
# Give the image to caffe
net.set_input_arrays(data/256, labels)
# Run the data through the network and get the results.
......@@ -67,7 +69,7 @@ out = net.forward()
# Print outputs, looping over minibatch. You should see that the network
# correctly classifies the image (it's the number 7).
for i in xrange(dlib_model.batch_size):
for i in xrange(dlib_model.input_batch_size):
print i, 'net final layer = ', out['fc1'][i]
print i, 'predicted number =', np.argmax(out['fc1'][i])
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment