dlib/examples/dnn_mmod_find_cars_ex.cpp

237 lines
11 KiB
C++
Raw Normal View History

2017-08-27 21:11:49 +08:00
// 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 vehicle detector using dlib. The
example loads a pretrained model and uses it to find the rear ends of cars in
2017-08-27 22:46:49 +08:00
an image. We will also visualize some of the detector's processing steps by
2017-08-27 21:11:49 +08:00
plotting various intermediate images on the screen. Viewing these can help
2017-08-27 22:46:49 +08:00
you understand how the detector works.
2017-08-27 21:11:49 +08:00
The model used by this example was trained by the dnn_mmod_train_find_cars_ex.cpp
example. Also, since this is a CNN, you really should use a GPU to get the
2017-09-16 10:16:00 +08:00
best execution speed. For instance, when run on a NVIDIA 1080ti, this detector
runs at 98fps when run on the provided test image. That's more than an order
of magnitude faster than when run on the CPU.
2017-08-27 21:11:49 +08:00
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.
2017-08-27 21:17:05 +08:00
You can also see some videos of this vehicle detector running on YouTube:
https://www.youtube.com/watch?v=4B3bzmxMAZU
https://www.youtube.com/watch?v=bP2SUo5vSlc
2017-08-27 21:11:49 +08:00
*/
2017-08-27 05:13:47 +08:00
#include <iostream>
#include <dlib/dnn.h>
2017-08-27 21:11:49 +08:00
#include <dlib/image_io.h>
2017-08-27 05:13:47 +08:00
#include <dlib/gui_widgets.h>
#include <dlib/image_processing.h>
using namespace std;
using namespace dlib;
2017-08-27 22:46:49 +08:00
// The rear view vehicle detector network
2017-08-27 05:13:47 +08:00
template <long num_filters, typename SUBNET> using con5d = con<num_filters,5,5,2,2,SUBNET>;
template <long num_filters, typename SUBNET> using con5 = con<num_filters,5,5,1,1,SUBNET>;
template <typename SUBNET> using downsampler = relu<affine<con5d<32, relu<affine<con5d<32, relu<affine<con5d<16,SUBNET>>>>>>>>>;
template <typename SUBNET> using rcon5 = relu<affine<con5<55,SUBNET>>>;
using net_type = loss_mmod<con<1,9,9,1,1,rcon5<rcon5<rcon5<downsampler<input_rgb_image_pyramid<pyramid_down<6>>>>>>>>;
// ----------------------------------------------------------------------------------------
int main() try
{
net_type net;
shape_predictor sp;
// You can get this file from http://dlib.net/files/mmod_rear_end_vehicle_detector.dat.bz2
// This network was produced by the dnn_mmod_train_find_cars_ex.cpp example program.
2017-09-17 03:35:37 +08:00
// As you can see, the file also includes a separately trained shape_predictor. To see
// a generic example of how to train those refer to train_shape_predictor_ex.cpp.
2017-08-27 05:13:47 +08:00
deserialize("mmod_rear_end_vehicle_detector.dat") >> net >> sp;
matrix<rgb_pixel> img;
load_image(img, "../mmod_cars_test_image.jpg");
image_window win;
win.set_image(img);
// Run the detector on the image and show us the output.
for (auto&& d : net(img))
{
// We use a shape_predictor to refine the exact shape and location of the detection
2017-08-27 22:46:49 +08:00
// box. This shape_predictor is trained to simply output the 4 corner points of
// the box. So all we do is make a rectangle that tightly contains those 4 points
// and that rectangle is our refined detection position.
2017-08-27 05:13:47 +08:00
auto fd = sp(img,d);
rectangle rect;
2017-08-27 21:23:57 +08:00
for (unsigned long j = 0; j < fd.num_parts(); ++j)
2017-08-27 05:13:47 +08:00
rect += fd.part(j);
win.add_overlay(rect, rgb_pixel(255,0,0));
}
cout << "Hit enter to view the intermediate processing steps" << endl;
cin.get();
2017-08-27 22:46:49 +08:00
// Now let's look at how the detector works. The high level processing steps look like:
2017-08-27 21:11:49 +08:00
// 1. Create an image pyramid and pack the pyramid into one big image. We call this
2017-08-28 06:18:00 +08:00
// image the "tiled pyramid".
2017-08-27 21:11:49 +08:00
// 2. Run the tiled pyramid image through the CNN. The CNN outputs a new image where
// bright pixels in the output image indicate the presence of cars.
2017-08-27 22:46:49 +08:00
// 3. Find pixels in the CNN's output image with a value > 0. Those locations are your
2017-08-27 21:11:49 +08:00
// preliminary car detections.
// 4. Perform non-maximum suppression on the preliminary detections to produce the
// final output.
//
// We will be plotting the images from steps 1 and 2 so you can visualize what's
2017-08-27 22:46:49 +08:00
// happening. For the CNN's output image, we will use the jet colormap so that "bright"
2017-08-27 21:11:49 +08:00
// outputs, i.e. pixels with big values, appear in red and "dim" outputs appear as a
// cold blue color. To do this we pick a range of CNN output values for the color
// mapping. The specific values don't matter. They are just selected to give a nice
// looking output image.
const float lower = -2.5;
const float upper = 0.0;
cout << "jet color mapping range: lower="<< lower << " upper="<< upper << endl;
2017-08-27 05:13:47 +08:00
2017-08-27 21:11:49 +08:00
// Create a tiled pyramid image and display it on the screen.
2017-08-27 05:13:47 +08:00
std::vector<rectangle> rects;
matrix<rgb_pixel> tiled_img;
2017-08-27 22:46:49 +08:00
// Get the type of pyramid the CNN used
using pyramid_type = std::remove_reference<decltype(input_layer(net))>::type::pyramid_type;
// And tell create_tiled_pyramid to create the pyramid using that pyramid type.
create_tiled_pyramid<pyramid_type>(img, tiled_img, rects,
input_layer(net).get_pyramid_padding(),
input_layer(net).get_pyramid_outer_padding());
2017-08-28 06:18:00 +08:00
image_window winpyr(tiled_img, "Tiled pyramid");
2017-08-27 05:13:47 +08:00
2017-08-27 22:46:49 +08:00
// This CNN detector represents a sliding window detector with 3 sliding windows. Each
// of the 3 windows has a different aspect ratio, allowing it to find vehicles which
// are either tall and skinny, squarish, or short and wide. The aspect ratio of a
// detection is determined by which channel in the output image triggers the detection.
// Here we are just going to max pool the channels together to get one final image for
// our display. In this image, a pixel will be bright if any of the sliding window
// detectors thinks there is a car at that location.
2017-08-27 05:13:47 +08:00
cout << "Number of channels in final tensor image: " << net.subnet().get_output().k() << endl;
matrix<float> network_output = image_plane(net.subnet().get_output(),0,0);
for (long k = 1; k < net.subnet().get_output().k(); ++k)
network_output = max_pointwise(network_output, image_plane(net.subnet().get_output(),0,k));
2017-08-27 22:46:49 +08:00
// We will also upsample the CNN's output image. The CNN we defined has an 8x
2017-08-27 21:11:49 +08:00
// downsampling layer at the beginning. In the code below we are going to overlay this
// CNN output image on top of the raw input image. To make that look nice it helps to
// upsample the CNN output image back to the same resolution as the input image, which
// we do here.
const double network_output_scale = img.nc()/(double)network_output.nc();
resize_image(network_output_scale, network_output);
2017-08-27 05:13:47 +08:00
2017-08-27 21:11:49 +08:00
// Display the network's output as a color image.
2017-08-27 05:13:47 +08:00
image_window win_output(jet(network_output, upper, lower), "Output tensor from the network");
2017-08-27 21:11:49 +08:00
// Also, overlay network_output on top of the tiled image pyramid and display it.
2017-08-27 22:46:49 +08:00
for (long r = 0; r < tiled_img.nr(); ++r)
2017-08-27 05:13:47 +08:00
{
2017-08-27 22:46:49 +08:00
for (long c = 0; c < tiled_img.nc(); ++c)
2017-08-27 05:13:47 +08:00
{
dpoint tmp(c,r);
tmp = input_tensor_to_output_tensor(net, tmp);
2017-08-27 21:11:49 +08:00
tmp = point(network_output_scale*tmp);
2017-08-27 05:13:47 +08:00
if (get_rect(network_output).contains(tmp))
{
float val = network_output(tmp.y(),tmp.x());
2017-08-27 21:11:49 +08:00
// alpha blend the network output pixel with the RGB image to make our
// overlay.
2017-08-27 05:13:47 +08:00
rgb_alpha_pixel p;
assign_pixel(p , colormap_jet(val,lower,upper));
p.alpha = 120;
2017-08-27 22:46:49 +08:00
assign_pixel(tiled_img(r,c), p);
2017-08-27 05:13:47 +08:00
}
}
}
2017-08-27 22:46:49 +08:00
// If you look at this image you can see that the vehicles have bright red blobs on
// them. That's the CNN saying "there is a car here!". You will also notice there is
// a certain scale at which it finds cars. They have to be not too big or too small,
// which is why we have an image pyramid. The pyramid allows us to find cars of all
// scales.
image_window win_pyr_overlay(tiled_img, "Detection scores on image pyramid");
2017-08-27 05:13:47 +08:00
2017-08-27 21:11:49 +08:00
// Finally, we can collapse the pyramid back into the original image. The CNN doesn't
// actually do this step, since it's enough to threshold the tiled pyramid image to get
// the detections. However, it makes a nice visualization and clearly indicates that
// the detector is firing for all the cars.
2017-08-27 22:46:49 +08:00
matrix<float> collapsed(img.nr(), img.nc());
2017-08-27 05:13:47 +08:00
resizable_tensor input_tensor;
input_layer(net).to_tensor(&img, &img+1, input_tensor);
2017-08-27 22:46:49 +08:00
for (long r = 0; r < collapsed.nr(); ++r)
2017-08-27 05:13:47 +08:00
{
2017-08-27 22:46:49 +08:00
for (long c = 0; c < collapsed.nc(); ++c)
2017-08-27 05:13:47 +08:00
{
2017-08-27 22:46:49 +08:00
// Loop over a bunch of scale values and look up what part of network_output
// corresponds to the point(c,r) in the original image, then take the max
// detection score over all the scales and save it at pixel point(c,r).
float max_score = -1e30;
2017-08-27 05:13:47 +08:00
for (double scale = 1; scale > 0.2; scale *= 5.0/6.0)
{
2017-08-27 21:11:49 +08:00
// Map from input image coordinates to tiled pyramid coordinates.
2017-08-27 05:13:47 +08:00
dpoint tmp = center(input_layer(net).image_space_to_tensor_space(input_tensor,scale, drectangle(dpoint(c,r))));
2017-08-27 21:11:49 +08:00
// Now map from pyramid coordinates to network_output coordinates.
tmp = point(network_output_scale*input_tensor_to_output_tensor(net, tmp));
2017-08-27 05:13:47 +08:00
if (get_rect(network_output).contains(tmp))
{
float val = network_output(tmp.y(),tmp.x());
2017-08-27 22:46:49 +08:00
if (val > max_score)
max_score = val;
2017-08-27 05:13:47 +08:00
}
}
2017-08-27 22:46:49 +08:00
collapsed(r,c) = max_score;
2017-08-27 05:13:47 +08:00
2017-08-27 22:46:49 +08:00
// Also blend the scores into the original input image so we can view it as
2017-08-27 05:13:47 +08:00
// an overlay on the cars.
rgb_alpha_pixel p;
2017-08-27 22:46:49 +08:00
assign_pixel(p , colormap_jet(max_score,lower,upper));
2017-08-27 05:13:47 +08:00
p.alpha = 120;
assign_pixel(img(r,c), p);
}
}
2017-08-27 22:46:49 +08:00
image_window win_collapsed(jet(collapsed, upper, lower), "Collapsed output tensor from the network");
image_window win_img_and_sal(img, "Collapsed detection scores on raw image");
2017-08-27 05:13:47 +08:00
cout << "Hit enter to end program" << endl;
cin.get();
}
catch(image_load_error& e)
{
cout << e.what() << endl;
cout << "The test image is located in the examples folder. So you should run this program from a sub folder so that the relative path is correct." << endl;
}
catch(serialization_error& e)
{
cout << e.what() << endl;
2017-09-17 02:10:09 +08:00
cout << "The correct model file can be obtained from: http://dlib.net/files/mmod_rear_end_vehicle_detector.dat.bz2" << endl;
2017-08-27 05:13:47 +08:00
}
catch(std::exception& e)
{
cout << e.what() << endl;
}