ZF/OpenFace
1
0
Fork 0
mirror of https://gitcode.com/gh_mirrors/ope/OpenFace.git synced 2026-05-14 11:17:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

More work on AU - WIP

Browse Source
This commit is contained in:
Tadas Baltrusaitis
2016-06-14 17:55:16 -04:00
parent 65e91da3a6
commit 146dcd5e5b
676 changed files with 2971 additions and 297 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
lib/local/FaceAnalyser/src/FaceAnalyser.cpp
View File

@@ -303,6 +303,59 @@ void FaceAnalyser::ExtractCurrentMedians(vector<cv::Mat>& hog_medians, vector<cv
}
}
std::pair<std::vector<std::pair<string, double>>, std::vector<std::pair<string, double>>> FaceAnalyser::PredictStaticAUs(const cv::Mat& frame, const LandmarkDetector::CLNF& clnf, bool visualise)
{
// First align the face
AlignFaceMask(aligned_face, frame, clnf, triangulation, true, align_scale, align_width, align_height);
// Extract HOG descriptor from the frame and convert it to a useable format
cv::Mat_<double> hog_descriptor;
Extract_FHOG_descriptor(hog_descriptor, aligned_face, this->num_hog_rows, this->num_hog_cols);
// Store the descriptor
hog_desc_frame = hog_descriptor;
cv::Vec3d curr_orient(clnf.params_global[1], clnf.params_global[2], clnf.params_global[3]);
int orientation_to_use = GetViewId(this->head_orientations, curr_orient);
// Geom descriptor and its median
geom_descriptor_frame = clnf.params_local.t();
// Stack with the actual feature point locations (without mean)
cv::Mat_<double> locs = clnf.pdm.princ_comp * geom_descriptor_frame.t();
cv::hconcat(locs.t(), geom_descriptor_frame.clone(), geom_descriptor_frame);
// First convert the face image to double representation as a row vector
cv::Mat_<uchar> aligned_face_cols(1, aligned_face.cols * aligned_face.rows * aligned_face.channels(), aligned_face.data, 1);
cv::Mat_<double> aligned_face_cols_double;
aligned_face_cols.convertTo(aligned_face_cols_double, CV_64F);
// Visualising the median HOG
if (visualise)
{
FaceAnalysis::Visualise_FHOG(hog_descriptor, num_hog_rows, num_hog_cols, hog_descriptor_visualisation);
}
// Perform AU prediction
auto AU_predictions_intensity = PredictCurrentAUs(orientation_to_use);
auto AU_predictions_occurence = PredictCurrentAUsClass(orientation_to_use);
// Make sure intensity is within range (0-5)
for (size_t au = 0; au < AU_predictions_intensity.size(); ++au)
{
if (AU_predictions_intensity[au].second < 0)
AU_predictions_intensity[au].second = 0;
if (AU_predictions_intensity[au].second > 5)
AU_predictions_intensity[au].second = 5;
}
return std::pair<std::vector<std::pair<std::string, double>>, std::vector<std::pair<std::string, double>>>(AU_predictions_intensity, AU_predictions_occurence);
}
void FaceAnalyser::AddNextFrame(const cv::Mat& frame, const LandmarkDetector::CLNF& clnf_model, double timestamp_seconds, bool online, bool visualise)
{
@@ -607,7 +660,6 @@ void FaceAnalyser::ExtractAllPredictionsOfflineReg(vector<std::pair<std::string,
confidences = this->confidences;
successes = this->valid_preds;
// TODO only if the video is long enough or there is enough range? Compare stdev of BP4D and this
for(auto au_iter = AU_predictions_reg_all_hist.begin(); au_iter != AU_predictions_reg_all_hist.end(); ++au_iter)
{
vector<double> au_good;
@@ -665,6 +717,28 @@ void FaceAnalyser::ExtractAllPredictionsOfflineReg(vector<std::pair<std::string,
}
}
// Perform some prediction smoothing
for (auto au_iter = au_predictions.begin(); au_iter != au_predictions.end(); ++au_iter)
{
string au_name = au_iter->first;
// Perform a moving average of 3 frames
int window_size = 3;
vector<double> au_vals_tmp = au_iter->second;
for (size_t i = (window_size - 1) / 2; i < au_iter->second.size() - (window_size - 1) / 2; ++i)
{
double sum = 0;
for (int w = -(window_size - 1) / 2; w < (window_size - 1) / 2; ++w)
{
sum += au_vals_tmp[i + w];
}
sum = sum / window_size;
au_iter->second[i] = sum;
}
}
}