NextCloud/dlib
1
0
Fork 0
mirror of https://github.com/davisking/dlib.git synced 2024-11-01 10:14:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

Added comments

Browse Source
This commit is contained in:
Davis King 2016-10-02 16:43:11 -04:00
parent d53d49ebb8
commit 56f4e19afa
2 changed files with 114 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

93
examples/dnn_mmod_dog_hipsterizer.cpp
View File

@ -1,3 +1,47 @@
// The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
/*
This example shows how to run a CNN based dog face detector using dlib. The
example loads a pretrained model and uses it to find dog faces in images.
We also use the dlib::shape_predictor to find the location of the eyes and
nose and then draw glasses and a mustache onto each dog found :)
Users who are just learning about dlib's deep learning API should read the
dnn_introduction_ex.cpp and dnn_introduction2_ex.cpp examples to learn how
the API works. For an introduction to the object detection method you
should read dnn_mmod_ex.cpp
TRAINING THE MODEL
Finally, users interested in how the dog face detector was trained should
read the dnn_mmod_ex.cpp example program. It should be noted that the
dog face detector used in this example uses a bigger training dataset and
larger CNN architecture than what is shown in dnn_mmod_ex.cpp, but
otherwise training is the same. If you compare the net_type statements
in this file and dnn_mmod_ex.cpp you will see that they are very similar
except that the number of parameters has been increased.
Additionally, the following training parameters were different during
training: The following lines in dnn_mmod_ex.cpp were changed from
mmod_options options(face_boxes_train, 40*40);
trainer.set_iterations_without_progress_threshold(300);
to the following when training the model used in this example:
mmod_options options(face_boxes_train, 80*80);
trainer.set_iterations_without_progress_threshold(8000);
Also, the random_cropper was left at its default settings, So we didn't
call these functions:
cropper.set_chip_dims(200, 200);
cropper.set_min_object_height(0.2);
The training data used to create the model is also available at
http://dlib.net/files/data/CU_dogs_fully_labeled.tar.gz
Lastly, the shape_predictor was trained with default settings except we
used the following non-default settings: cascade depth=20, tree
depth=5, padding=0.2
*/
#include <iostream>
@ -10,30 +54,6 @@
using namespace std;
using namespace dlib;
/*
Training differences with dnn_mmod_ex.cpp
A slightly bigger network architecture. Also, to train you must replace the affine layers with bn_con layers.
mmod_options options(training_labels, 80*80);
instead of
mmod_options options(face_boxes_train, 40*40);
trainer.set_iterations_without_progress_threshold(8000);
instead of
trainer.set_iterations_without_progress_threshold(300);
random cropper was left at its default settings, So we didn't call these functions:
cropper.set_chip_dims(200, 200);
cropper.set_min_object_height(0.2);
// shape predictor was trained with these settings: tree cascade depth=20, tree depth=5, padding=0.2
*/
// ----------------------------------------------------------------------------------------
template <long num_filters, typename SUBNET> using con5d = con<num_filters,5,5,2,2,SUBNET>;
@ -46,20 +66,19 @@ using net_type = loss_mmod<con<1,9,9,1,1,rcon5<rcon5<rcon5<downsampler<input_rgb
// ----------------------------------------------------------------------------------------
int main(int argc, char** argv) try
{
if (argc < 3)
{
cout << "Give the path to the examples/faces directory as the argument to this" << endl;
cout << "program. For example, if you are in the examples folder then execute " << endl;
cout << "this program by running: " << endl;
cout << " ./fhog_object_detector_ex faces" << endl;
cout << endl;
cout << "Call this program like this:" << endl;
cout << "./dnn_mmod_dog_hipsterizer mmod_dog_hipsterizer.dat faces/dogs.jpg" << endl;
cout << "\nYou can get the mmod_dog_hipsterizer.dat file from:\n";
cout << "http://dlib.net/files/mmod_dog_hipsterizer.dat.bz2" << endl;
return 0;
}
// load the models as well as glasses and mustache.
net_type net;
shape_predictor sp;
matrix<rgb_alpha_pixel> glasses, mustache;
@ -67,11 +86,12 @@ int main(int argc, char** argv) try
pyramid_up(glasses);
pyramid_up(mustache);
// right eye (59,35), left eye (176,36)
image_window win1(glasses);
image_window win2(mustache);
image_window win_wireframe, win_hipster;
// Now process each image, find dogs, and hipsterize them by drawing glasses and a
// mustache on each dog :)
for (int i = 2; i < argc; ++i)
{
matrix<rgb_pixel> img;
@ -84,9 +104,12 @@ int main(int argc, char** argv) try
auto dets = net(img);
win_wireframe.clear_overlay();
win_wireframe.set_image(img);
// We will also draw a wireframe on each dog's face so you can see where the
// shape_predictor is identifying face landmarks.
std::vector<image_window::overlay_line> lines;
for (auto&& d : dets)
{
// get the landmarks for this dog's face
auto shape = sp(img, d.rect);
const rgb_pixel color(0,255,0);
@ -97,9 +120,11 @@ int main(int argc, char** argv) try
auto rear = shape.part(4);
auto reye = shape.part(5);
// The locations of the left and right ends of the mustache.
auto lmustache = 1.3*(leye-reye)/2 + nose;
auto rmustache = 1.3*(reye-leye)/2 + nose;
// Draw the glasses onto the image.
std::vector<point> from = {2*point(176,36), 2*point(59,35)}, to = {leye, reye};
auto tform = find_similarity_transform(from, to);
for (long r = 0; r < glasses.nr(); ++r)
@ -111,6 +136,8 @@ int main(int argc, char** argv) try
assign_pixel(img(p.y(),p.x()), glasses(r,c));
}
}
// Draw the mustache onto the image right under the dog's nose.
auto mrect = get_rect(mustache);
from = {mrect.tl_corner(), mrect.tr_corner()};
to = {rmustache, lmustache};
@ -126,6 +153,7 @@ int main(int argc, char** argv) try
}
// Record the lines needed for the face wire frame.
lines.push_back(image_window::overlay_line(leye, nose, color));
lines.push_back(image_window::overlay_line(nose, reye, color));
lines.push_back(image_window::overlay_line(reye, leye, color));
@ -138,6 +166,7 @@ int main(int argc, char** argv) try
win_wireframe.add_overlay(lines);
win_hipster.set_image(img);
cout << "Hit enter to process the next image." << endl;
cin.get();
}
}

85
examples/dnn_mmod_face_detection_ex.cpp
View File

@ -1,3 +1,45 @@
// The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
/*
This example shows how to run a CNN based face detector using dlib. The
example loads a pretrained model and uses it to find faces in images. The
CNN model is much more accurate than the HOG based model shown in the
face_detection_ex.cpp example, but takes much more computational power to
run, and is meant to be executed on a GPU to attain reasonable speed. For
example, on a NVIDIA Titan X GPU, this example program processes images at
about the same speed as face_detection_ex.cpp.
Also, users who are just learning about dlib's deep learning API should read
the dnn_introduction_ex.cpp and dnn_introduction2_ex.cpp examples to learn
how the API works. For an introduction to the object detection method you
should read dnn_mmod_ex.cpp
TRAINING THE MODEL
Finally, users interested in how the face detector was trained should
read the dnn_mmod_ex.cpp example program. It should be noted that the
face detector used in this example uses a bigger training dataset and
larger CNN architecture than what is shown in dnn_mmod_ex.cpp, but
otherwise training is the same. If you compare the net_type statements
in this file and dnn_mmod_ex.cpp you will see that they are very similar
except that the number of parameters has been increased.
Additionally, the following training parameters were different during
training: The following lines in dnn_mmod_ex.cpp were changed from
mmod_options options(face_boxes_train, 40*40);
trainer.set_iterations_without_progress_threshold(300);
to the following when training the model used in this example:
mmod_options options(face_boxes_train, 80*80);
trainer.set_iterations_without_progress_threshold(8000);
Also, the random_cropper was left at its default settings, So we didn't
call these functions:
cropper.set_chip_dims(200, 200);
cropper.set_min_object_height(0.2);
The training data used to create the model is also available at
http://dlib.net/files/data/dlib_face_detection_dataset-2016-09-30.tar.gz
*/
#include <iostream>
@ -10,26 +52,6 @@
using namespace std;
using namespace dlib;
/*
Training differences with dnn_mmod_ex.cpp
A slightly bigger network architecture. Also, to train you must replace the affine layers with bn_con layers.
mmod_options options(training_labels, 80*80);
instead of
mmod_options options(face_boxes_train, 40*40);
trainer.set_iterations_without_progress_threshold(8000);
instead of
trainer.set_iterations_without_progress_threshold(300);
random cropper was left at its default settings, So we didn't call these functions:
cropper.set_chip_dims(200, 200);
cropper.set_min_object_height(0.2);
*/
// ----------------------------------------------------------------------------------------
template <long num_filters, typename SUBNET> using con5d = con<num_filters,5,5,2,2,SUBNET>;
@ -45,13 +67,12 @@ using net_type = loss_mmod<con<1,9,9,1,1,rcon5<rcon5<rcon5<downsampler<input_rgb
int main(int argc, char** argv) try
{
if (argc < 3)
if (argc == 1)
{
cout << "Give the path to the examples/faces directory as the argument to this" << endl;
cout << "program. For example, if you are in the examples folder then execute " << endl;
cout << "this program by running: " << endl;
cout << " ./fhog_object_detector_ex faces" << endl;
cout << endl;
cout << "Call this program like this:" << endl;
cout << "./dnn_mmod_face_detection_ex mmod_human_face_detector.dat faces/*.jpg" << endl;
cout << "\nYou can get the mmod_human_face_detector.dat file from:\n";
cout << "http://dlib.net/files/mmod_human_face_detector.dat.bz2" << endl;
return 0;
}
@ -71,15 +92,17 @@ int main(int argc, char** argv) try
pyramid_up(img);
// Note that you can process a bunch of images in a std::vector at once and it runs
// faster, since this will form mini-batches of images and therefore get better
// parallelism out of your GPU hardware. However, all the images must be the same
// size. To avoid this requirement on images being the same size we process them
// individually in this example.
// much faster, since this will form mini-batches of images and therefore get
// better parallelism out of your GPU hardware. However, all the images must be
// the same size. To avoid this requirement on images being the same size we
// process them individually in this example.
auto dets = net(img);
win.clear_overlay();
win.set_image(img);
for (auto&& d : dets)
win.add_overlay(d);
cout << "Hit enter to process the next image." << endl;
cin.get();
}
}
@ -89,5 +112,3 @@ catch(std::exception& e)
}