Added the program that made the resnet model.

examples/CMakeLists.txt

@@ -35,6 +35,7 @@ if (COMPILER_CAN_DO_CPP_11)
    add_example(dnn_mnist_advanced_ex)
    add_example(dnn_inception_ex)
    add_example(dnn_imagenet_ex)
+   add_example(dnn_imagenet_train_ex)
 endif()
 
 #here we apply our macros 

examples/dnn_imagenet_ex.cpp

 // The contents of this file are in the public domain. See LICENSE_FOR_EXAMPLE_PROGRAMS.txt
 /*
-    This example shows how to classify an image into one of the 1000 imagenet clategories
-    using the deep learning tools from the dlib C++ Library.  We will use the pretrained
-    ResNet34 model available on the dlib website.
+    This example shows how to classify an image into one of the 1000 imagenet
+    categories using the deep learning tools from the dlib C++ Library.  We will
+    use the pretrained ResNet34 model available on the dlib website.
 
-    The ResNet34 model is from Deep Residual Learning for Image Recognition by He, Zhang,
-    Ren, and Sun.  
+    The ResNet34 architecture is from the paper Deep Residual Learning for Image
+    Recognition by He, Zhang, Ren, and Sun.  The model file that comes with dlib
+    was trained using the dnn_imagenet_train_ex.cpp program on a Titan X for
+    about 2 weeks.  This pretrained model has a top5 error of 7.572% on the 2012
+    imagenet validation dataset.
+
+    For an introduction to dlib's DNN module read the dnn_mnist_ex.cpp and
+    dnn_mnist_advanced_ex.cpp example programs.
 
     
-    These tools will use CUDA and cuDNN to drastically accelerate network
-    training and testing.  CMake should automatically find them if they are
-    installed and configure things appropriately.  If not, the program will
+    Finally, these tools will use CUDA and cuDNN to drastically accelerate
+    network training and testing.  CMake should automatically find them if they
+    are installed and configure things appropriately.  If not, the program will
     still run but will be much slower to execute.
 */
 
@@ -27,6 +33,7 @@ using namespace dlib;
  
 // ----------------------------------------------------------------------------------------
 
+// This block of statements defines the resnet-34 network
 
 template <template <int,template<typename>class,int,typename> class block, int N, template<typename>class BN, typename SUBNET>
 using residual = add_prev1<block<N,BN,1,tag1<SUBNET>>>;
@@ -41,14 +48,14 @@ template <int N, typename SUBNET> using ares      = relu<residual<block,N,affine
 template <int N, typename SUBNET> using ares_down = relu<residual_down<block,N,affine,SUBNET>>;
 
 
-typedef loss_multiclass_log<fc<1000,avg_pool_everything<
+using anet_type = loss_multiclass_log<fc<1000,avg_pool_everything<
                             ares<512,ares<512,ares_down<512,
                             ares<256,ares<256,ares<256,ares<256,ares<256,ares_down<256,
                             ares<128,ares<128,ares<128,ares_down<128,
                             ares<64,ares<64,ares<64,
                             max_pool<3,3,2,2,relu<affine<con<64,7,7,2,2,
                             input_rgb_image_sized<227>
-                            >>>>>>>>>>>>>>>>>>>>>>> anet_type;
+                            >>>>>>>>>>>>>>>>>>>>>>>;
 
 // ----------------------------------------------------------------------------------------
 
@@ -101,11 +108,17 @@ void randomly_crop_images (
 
 int main(int argc, char** argv) try
 {
+    if (argc == 1)
+    {
+        cout << "Give this program image files as command line arguments.\n" << endl;
+        cout << "You will also need a copy of the file resnet34_1000_imagenet_classifier.dnn which" << endl;
+        cout << "is available at http://dlib.net/files/resnet34_1000_imagenet_classifier.dnn.bz2" << endl;
+        cout << endl;
+        return 1;
+    }
+
     std::vector<string> labels;
     anet_type net;
-    // Get this file from http://dlib.net/files/resnet34_1000_imagenet_classifier.dnn.bz2
-    // This pretrained model has a top5 error of 7.572% on the 2012 imagenet validation
-    // dataset.
     deserialize("resnet34_1000_imagenet_classifier.dnn") >> net >> labels;
 
 
@@ -118,13 +131,15 @@ int main(int argc, char** argv) try
     dlib::rand rnd;
     image_window win;
 
-    // read images from the command prompt and print the top 5 best labels.
+    // read images from the command prompt and print the top 5 best labels for each.
     for (int i = 1; i < argc; ++i)
     {
         load_image(img, argv[i]);
         const int num_crops = 16;
+        // Grab 16 random crops from the image.  We will run all of them through the
+        // network and average the results.
         randomly_crop_images(img, images, rnd, num_crops);
-
+        // p(i) == the probability the image contains object of class i.
         matrix<float,1,1000> p = sum_rows(mat(snet(images.begin(), images.end())))/num_crops;
 
         win.set_image(img);
@@ -135,6 +150,7 @@ int main(int argc, char** argv) try
             p(predicted_label) = 0;
         }
 
+        cout << "Hit enter to process the next image";
         cin.get();
     }