Files
OpenFace/lib/local/LandmarkDetector/src/LandmarkDetectionValidator.cpp

554 lines
17 KiB
C++

///////////////////////////////////////////////////////////////////////////////
// Copyright (C) 2017, Carnegie Mellon University and University of Cambridge,
// all rights reserved.
//
// ACADEMIC OR NON-PROFIT ORGANIZATION NONCOMMERCIAL RESEARCH USE ONLY
//
// BY USING OR DOWNLOADING THE SOFTWARE, YOU ARE AGREEING TO THE TERMS OF THIS LICENSE AGREEMENT.
// IF YOU DO NOT AGREE WITH THESE TERMS, YOU MAY NOT USE OR DOWNLOAD THE SOFTWARE.
//
// License can be found in OpenFace-license.txt
//
// * Any publications arising from the use of this software, including but
// not limited to academic journal and conference publications, technical
// reports and manuals, must cite at least one of the following works:
//
// OpenFace 2.0: Facial Behavior Analysis Toolkit
// Tadas Baltrušaitis, Amir Zadeh, Yao Chong Lim, and Louis-Philippe Morency
// in IEEE International Conference on Automatic Face and Gesture Recognition, 2018
//
// Convolutional experts constrained local model for facial landmark detection.
// A. Zadeh, T. Baltrušaitis, and Louis-Philippe Morency,
// in Computer Vision and Pattern Recognition Workshops, 2017.
//
// Rendering of Eyes for Eye-Shape Registration and Gaze Estimation
// Erroll Wood, Tadas Baltrušaitis, Xucong Zhang, Yusuke Sugano, Peter Robinson, and Andreas Bulling
// in IEEE International. Conference on Computer Vision (ICCV), 2015
//
// Cross-dataset learning and person-specific normalisation for automatic Action Unit detection
// Tadas Baltrušaitis, Marwa Mahmoud, and Peter Robinson
// in Facial Expression Recognition and Analysis Challenge,
// IEEE International Conference on Automatic Face and Gesture Recognition, 2015
//
///////////////////////////////////////////////////////////////////////////////
#include "stdafx.h"
#include "LandmarkDetectionValidator.h"
// OpenCV includes
#include <opencv2/core/core.hpp>
#include <opencv2/imgproc.hpp>
// TBB includes
#include <tbb/tbb.h>
// System includes
#include <fstream>
// Math includes
#define _USE_MATH_DEFINES
#include <cmath>
#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif
// Local includes
#include "LandmarkDetectorUtils.h"
#include "CNN_utils.h"
using namespace LandmarkDetector;
// Copy constructor
DetectionValidator::DetectionValidator(const DetectionValidator& other) : orientations(other.orientations), paws(other.paws),
cnn_subsampling_layers(other.cnn_subsampling_layers), cnn_layer_types(other.cnn_layer_types), cnn_convolutional_layers_im2col_precomp(other.cnn_convolutional_layers_im2col_precomp),
cnn_convolutional_layers_weights(other.cnn_convolutional_layers_weights)
{
this->cnn_convolutional_layers.resize(other.cnn_convolutional_layers.size());
for (size_t v = 0; v < other.cnn_convolutional_layers.size(); ++v)
{
this->cnn_convolutional_layers[v].resize(other.cnn_convolutional_layers[v].size());
for (size_t l = 0; l < other.cnn_convolutional_layers[v].size(); ++l)
{
this->cnn_convolutional_layers[v][l].resize(other.cnn_convolutional_layers[v][l].size());
for (size_t i = 0; i < other.cnn_convolutional_layers[v][l].size(); ++i)
{
this->cnn_convolutional_layers[v][l][i].resize(other.cnn_convolutional_layers[v][l][i].size());
for (size_t k = 0; k < other.cnn_convolutional_layers[v][l][i].size(); ++k)
{
// Make sure the matrix is copied.
this->cnn_convolutional_layers[v][l][i][k] = other.cnn_convolutional_layers[v][l][i][k].clone();
}
}
}
}
this->cnn_fully_connected_layers_weights.resize(other.cnn_fully_connected_layers_weights.size());
for (size_t v = 0; v < other.cnn_fully_connected_layers_weights.size(); ++v)
{
this->cnn_fully_connected_layers_weights[v].resize(other.cnn_fully_connected_layers_weights[v].size());
for (size_t l = 0; l < other.cnn_fully_connected_layers_weights[v].size(); ++l)
{
// Make sure the matrix is copied.
this->cnn_fully_connected_layers_weights[v][l] = other.cnn_fully_connected_layers_weights[v][l].clone();
}
}
this->cnn_fully_connected_layers_biases.resize(other.cnn_fully_connected_layers_biases.size());
for (size_t v = 0; v < other.cnn_fully_connected_layers_biases.size(); ++v)
{
this->cnn_fully_connected_layers_biases[v].resize(other.cnn_fully_connected_layers_biases[v].size());
for (size_t l = 0; l < other.cnn_fully_connected_layers_biases[v].size(); ++l)
{
// Make sure the matrix is copied.
this->cnn_fully_connected_layers_biases[v][l] = other.cnn_fully_connected_layers_biases[v][l].clone();
}
}
this->mean_images.resize(other.mean_images.size());
for (size_t i = 0; i < other.mean_images.size(); ++i)
{
// Make sure the matrix is copied.
this->mean_images[i] = other.mean_images[i].clone();
}
this->standard_deviations.resize(other.standard_deviations.size());
for (size_t i = 0; i < other.standard_deviations.size(); ++i)
{
// Make sure the matrix is copied.
this->standard_deviations[i] = other.standard_deviations[i].clone();
}
}
//===========================================================================
// Read in the landmark detection validation module
void DetectionValidator::Read(string location)
{
ifstream detection_validator_stream (location, ios::in|ios::binary);
if (detection_validator_stream.is_open())
{
detection_validator_stream.seekg (0, ios::beg);
// Read validator type
int validator_type;
detection_validator_stream.read ((char*)&validator_type, 4);
if (validator_type != 3)
{
cout << "ERROR: Using old face validator, no longer supported" << endl;
}
// Read the number of views (orientations) within the validator
int n;
detection_validator_stream.read ((char*)&n, 4);
orientations.resize(n);
for(int i = 0; i < n; i++)
{
cv::Mat_<double> orientation_tmp;
LandmarkDetector::ReadMatBin(detection_validator_stream, orientation_tmp);
orientations[i] = cv::Vec3d(orientation_tmp.at<double>(0), orientation_tmp.at<double>(1), orientation_tmp.at<double>(2));
// Convert from degrees to radians
orientations[i] = orientations[i] * M_PI / 180.0;
}
// Initialise the piece-wise affine warps, biases and weights
paws.resize(n);
cnn_convolutional_layers_weights.resize(n);
cnn_convolutional_layers_im2col_precomp.resize(n);
cnn_convolutional_layers.resize(n);
cnn_fully_connected_layers_weights.resize(n);
cnn_layer_types.resize(n);
cnn_fully_connected_layers_biases.resize(n);
// Initialise the normalisation terms
mean_images.resize(n);
standard_deviations.resize(n);
// Read in the validators for each of the views
for(int i = 0; i < n; i++)
{
// Read in the mean images
cv::Mat_<double> mean_img;
LandmarkDetector::ReadMatBin(detection_validator_stream, mean_img);
mean_img.convertTo(mean_images[i], CV_32F);
mean_images[i] = mean_images[i].t();
cv::Mat_<double> std_dev;
LandmarkDetector::ReadMatBin(detection_validator_stream, std_dev);
std_dev.convertTo(standard_deviations[i], CV_32F);
standard_deviations[i] = standard_deviations[i].t();
// Model specifics
if (validator_type == 3)
{
int network_depth;
detection_validator_stream.read((char*)&network_depth, 4);
cnn_layer_types[i].resize(network_depth);
for (int layer = 0; layer < network_depth; ++layer)
{
int layer_type;
detection_validator_stream.read((char*)&layer_type, 4);
cnn_layer_types[i][layer] = layer_type;
// convolutional
if (layer_type == 0)
{
// Read the number of input maps
int num_in_maps;
detection_validator_stream.read((char*)&num_in_maps, 4);
// Read the number of kernels for each input map
int num_kernels;
detection_validator_stream.read((char*)&num_kernels, 4);
vector<vector<cv::Mat_<float> > > kernels;
kernels.resize(num_in_maps);
vector<float> biases;
for (int k = 0; k < num_kernels; ++k)
{
float bias;
detection_validator_stream.read((char*)&bias, 4);
biases.push_back(bias);
}
// For every input map
for (int in = 0; in < num_in_maps; ++in)
{
kernels[in].resize(num_kernels);
// For every kernel on that input map
for (int k = 0; k < num_kernels; ++k)
{
ReadMatBin(detection_validator_stream, kernels[in][k]);
}
}
cnn_convolutional_layers[i].push_back(kernels);
// Rearrange the kernels for faster inference with FFT
vector<vector<cv::Mat_<float> > > kernels_rearr;
kernels_rearr.resize(num_kernels);
// Fill up the rearranged layer
for (int k = 0; k < num_kernels; ++k)
{
for (int in = 0; in < num_in_maps; ++in)
{
kernels_rearr[k].push_back(kernels[in][k]);
}
}
// Rearrange the flattened kernels into weight matrices for direct convolution computation
cv::Mat_<float> weight_matrix(num_in_maps * kernels_rearr[0][0].rows * kernels_rearr[0][0].cols, num_kernels);
for (int k = 0; k < num_kernels; ++k)
{
for (int i = 0; i < num_in_maps; ++i)
{
// Flatten the kernel
cv::Mat_<float> k_flat = kernels_rearr[k][i].t();
k_flat = k_flat.reshape(0, 1).t();
k_flat.copyTo(weight_matrix(cv::Rect(k, i * kernels_rearr[0][0].rows * kernels_rearr[0][0].cols, 1, kernels_rearr[0][0].rows * kernels_rearr[0][0].cols)));
}
}
// Transpose the weight matrix for more convenient computation
weight_matrix = weight_matrix.t();
// Add a bias term to the weight matrix for efficiency
cv::Mat_<float> W(weight_matrix.rows, weight_matrix.cols + 1, 1.0);
for (int k = 0; k < weight_matrix.rows; ++k)
{
W.at<float>(k, weight_matrix.cols) = biases[k];
}
weight_matrix.copyTo(W(cv::Rect(0, 0, weight_matrix.cols, weight_matrix.rows)));
cnn_convolutional_layers_weights[i].push_back(W.t());
cnn_convolutional_layers_im2col_precomp[i].push_back(cv::Mat_<float>());
}
else if (layer_type == 2)
{
cv::Mat_<float> biases;
ReadMatBin(detection_validator_stream, biases);
cnn_fully_connected_layers_biases[i].push_back(biases);
// Fully connected layer
cv::Mat_<float> weights;
ReadMatBin(detection_validator_stream, weights);
cnn_fully_connected_layers_weights[i].push_back(weights);
}
}
}
// Read in the piece-wise affine warps
paws[i].Read(detection_validator_stream);
}
}
else
{
cout << "WARNING: Can't find the Face checker location" << endl;
}
}
//===========================================================================
// Check if the fitting actually succeeded
float DetectionValidator::Check(const cv::Vec3d& orientation, const cv::Mat_<uchar>& intensity_img, cv::Mat_<float>& detected_landmarks)
{
int id = GetViewId(orientation);
// The warped (cropped) image, corresponding to a face lying withing the detected lanmarks
cv::Mat_<float> warped;
// First only use the ROI of the image of interest
cv::Mat_<float> detected_landmarks_local = detected_landmarks.clone();
float min_x_f, max_x_f, min_y_f, max_y_f;
ExtractBoundingBox(detected_landmarks_local, min_x_f, max_x_f, min_y_f, max_y_f);
cv::Mat_<float> xs = detected_landmarks_local(cv::Rect(0, 0, 1, detected_landmarks.rows / 2));
cv::Mat_<float> ys = detected_landmarks_local(cv::Rect(0, detected_landmarks.rows / 2, 1, detected_landmarks.rows / 2));
// Picking the ROI (some extra space for bilinear interpolation)
int min_x = (int)(min_x_f - 3.0f);
int max_x = (int)(max_x_f + 3.0f);
int min_y = (int)(min_y_f - 3.0f);
int max_y = (int)(max_y_f + 3.0f);
if (min_x < 0) min_x = 0;
if (min_y < 0) min_y = 0;
if (max_x > intensity_img.cols - 1) max_x = intensity_img.cols - 1;
if (max_y > intensity_img.rows - 1) max_y = intensity_img.rows - 1;
xs = xs - min_x;
ys = ys - min_y;
// If the ROI is non existent return failure (this could happen if all landmarks are outside of the image)
if (max_x - min_x <= 1 || max_y - min_y <= 1)
{
return 0.0f;
}
cv::Mat_<float> intensity_img_float_local;
intensity_img(cv::Rect(min_x, min_y, max_x - min_x, max_y - min_y)).convertTo(intensity_img_float_local, CV_32F);
// the piece-wise affine image warping
paws[id].Warp(intensity_img_float_local, warped, detected_landmarks_local);
// The actual validation step
double dec = CheckCNN(warped, id);
// Convert it to a more interpretable signal (0 low confidence, 1 high confidence)
dec = 0.5 * (1.0 - dec);
return (float)dec;
}
double DetectionValidator::CheckCNN(const cv::Mat_<float>& warped_img, int view_id)
{
cv::Mat_<float> feature_vec;
NormaliseWarpedToVector(warped_img, feature_vec, view_id);
// Create a normalised image from the crop vector
cv::Mat_<float> img(warped_img.size(), 0.0);
img = img.t();
cv::Mat mask = paws[view_id].pixel_mask.t();
cv::MatIterator_<uchar> mask_it = mask.begin<uchar>();
cv::MatIterator_<float> feature_it = feature_vec.begin();
cv::MatIterator_<float> img_it = img.begin();
int wInt = img.cols;
int hInt = img.rows;
for (int i = 0; i < wInt; ++i)
{
for (int j = 0; j < hInt; ++j, ++mask_it, ++img_it)
{
// if is within mask
if (*mask_it)
{
// assign the feature to image if it is within the mask
*img_it = (float)*feature_it++;
}
}
}
img = img.t();
int cnn_layer = 0;
int fully_connected_layer = 0;
vector<cv::Mat_<float> > input_maps;
input_maps.push_back(img);
vector<cv::Mat_<float> > outputs;
for (size_t layer = 0; layer < cnn_layer_types[view_id].size(); ++layer)
{
// Determine layer type
int layer_type = cnn_layer_types[view_id][layer];
// Convolutional layer
if (layer_type == 0)
{
convolution_direct_blas(outputs, input_maps, cnn_convolutional_layers_weights[view_id][cnn_layer], cnn_convolutional_layers[view_id][cnn_layer][0][0].rows, cnn_convolutional_layers[view_id][cnn_layer][0][0].cols, cnn_convolutional_layers_im2col_precomp[view_id][cnn_layer]);
cnn_layer++;
}
if (layer_type == 1)
{
max_pooling(outputs, input_maps, 2, 2, 2, 2);
}
if (layer_type == 2)
{
fully_connected(outputs, input_maps, cnn_fully_connected_layers_weights[view_id][fully_connected_layer].t(), cnn_fully_connected_layers_biases[view_id][fully_connected_layer]);
fully_connected_layer++;
}
if (layer_type == 3) // ReLU
{
outputs.clear();
for (size_t k = 0; k < input_maps.size(); ++k)
{
// Apply the ReLU
cv::threshold(input_maps[k], input_maps[k], 0, 0, cv::THRESH_TOZERO);
outputs.push_back(input_maps[k]);
}
}
if (layer_type == 4)
{
outputs.clear();
for (size_t k = 0; k < input_maps.size(); ++k)
{
// Apply the sigmoid
cv::exp(-input_maps[k], input_maps[k]);
input_maps[k] = 1.0 / (1.0 + input_maps[k]);
outputs.push_back(input_maps[k]);
}
}
// Set the outputs of this layer to inputs of the next
input_maps = outputs;
}
// Convert the class label to a continuous value
double max_val = 0;
cv::Point max_loc;
cv::minMaxLoc(outputs[0].t(), 0, &max_val, 0, &max_loc);
int max_idx = max_loc.y;
double max = 1;
double min = -1;
double bins = (double)outputs[0].cols;
// Unquantizing the softmax layer to continuous value
double step_size = (max - min) / bins; // This should be saved somewhere
double unquantized = min + step_size / 2.0 + max_idx * step_size;
return unquantized;
}
void DetectionValidator::NormaliseWarpedToVector(const cv::Mat_<float>& warped_img, cv::Mat_<float>& feature_vec, int view_id)
{
cv::Mat_<float> warped_t = warped_img.t();
// the vector to be filled with paw values
cv::MatIterator_<float> vp;
cv::MatIterator_<float> cp;
cv::Mat_<float> vec(paws[view_id].number_of_pixels,1);
vp = vec.begin();
cp = warped_t.begin();
int wInt = warped_img.cols;
int hInt = warped_img.rows;
// the mask indicating if point is within or outside the face region
cv::Mat maskT = paws[view_id].pixel_mask.t();
cv::MatIterator_<uchar> mp = maskT.begin<uchar>();
for(int i=0; i < wInt; ++i)
{
for(int j=0; j < hInt; ++j, ++mp, ++cp)
{
// if is within mask
if(*mp)
{
*vp++ = *cp;
}
}
}
// Local normalisation
cv::Scalar mean;
cv::Scalar std;
cv::meanStdDev(vec, mean, std);
// subtract the mean image
vec -= mean[0];
// Normalise the image
if(std[0] == 0)
{
std[0] = 1;
}
vec /= std[0];
// Global normalisation
feature_vec = (vec - mean_images[view_id]) / standard_deviations[view_id];
}
// Getting the closest view center based on orientation
int DetectionValidator::GetViewId(const cv::Vec3d& orientation) const
{
int id = 0;
double dbest = -1.0;
for(size_t i = 0; i < this->orientations.size(); i++)
{
// Distance to current view
double d = cv::norm(orientation, this->orientations[i]);
if(i == 0 || d < dbest)
{
dbest = d;
id = i;
}
}
return id;
}