ZF/OpenFace
1
0
Fork 0
mirror of https://gitcode.com/gh_mirrors/ope/OpenFace.git synced 2026-02-18 13:30:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

Moving to AVX instead of AVX2, compiler warning fixes. Losing dependency on OpenCV C style bindings for affine warping.

Browse Source
This commit is contained in:
Tadas Baltrusaitis
2018-05-07 21:54:52 +01:00
parent c376351145
commit e6849c6b52
182 changed files with 155 additions and 232 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
lib/local/LandmarkDetector/src/CEN_patch_expert.cpp
View File

@@ -77,7 +77,7 @@
using namespace LandmarkDetector;
// Copy constructor
CEN_patch_expert::CEN_patch_expert(const CEN_patch_expert& other) : confidence(other.confidence), width(other.width), height(other.height)
CEN_patch_expert::CEN_patch_expert(const CEN_patch_expert& other) : confidence(other.confidence), width_support(other.width_support), height_support(other.height_support)
{
// Copy the layer weights in a deep way
@@ -106,8 +106,8 @@ void CEN_patch_expert::Read(ifstream &stream)
// the number of neurons for this patch
int num_layers;
stream.read((char*)&width, 4);
stream.read((char*)&height, 4);
stream.read((char*)&width_support, 4);
stream.read((char*)&height_support, 4);
stream.read((char*)&num_layers, 4);
if (num_layers == 0)
@@ -221,11 +221,11 @@ void im2colBias(const cv::Mat_<float>& input, int width, int height, cv::Mat_<fl
void CEN_patch_expert::Response(const cv::Mat_<float> &area_of_interest, cv::Mat_<float> &response)
{
int response_height = area_of_interest.rows - height + 1;
int response_width = area_of_interest.cols - width + 1;
int response_height = area_of_interest.rows - height_support + 1;
int response_width = area_of_interest.cols - width_support + 1;
cv::Mat_<float> input_col;
im2colBias(area_of_interest, width, height, input_col);
im2colBias(area_of_interest, width_support, height_support, input_col);
// Mean and standard deviation normalization
contrastNorm(input_col, response);
@@ -246,8 +246,7 @@ void CEN_patch_expert::Response(const cv::Mat_<float> &area_of_interest, cv::Mat
// Perform matrix multiplication in OpenBLAS (fortran call)
float alpha1 = 1.0;
float beta1 = 0.0;
char *nT = "N";
sgemm_(nT, nT, &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
sgemm_("N", "N", &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
// The above is a faster version of this, by calling the fortran version directly
//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);
@@ -507,11 +506,11 @@ void LandmarkDetector::interpolationMatrix(cv::Mat_<float>& mapMatrix, int respo
void CEN_patch_expert::ResponseSparse(const cv::Mat_<float> &area_of_interest, cv::Mat_<float> &response, cv::Mat_<float>& mapMatrix, cv::Mat_<float>& im2col_prealloc)
{
int response_height = area_of_interest.rows - height + 1;
int response_width = area_of_interest.cols - width + 1;
int response_height = area_of_interest.rows - height_support + 1;
int response_width = area_of_interest.cols - width_support + 1;
// Extract im2col but in a sparse way and contrast normalize
im2colBiasSparseContrastNorm(area_of_interest, width, height, im2col_prealloc);
im2colBiasSparseContrastNorm(area_of_interest, width_support, height_support, im2col_prealloc);
response = im2col_prealloc.t();
for (size_t layer = 0; layer < activation_function.size(); ++layer)
@@ -529,8 +528,7 @@ void CEN_patch_expert::ResponseSparse(const cv::Mat_<float> &area_of_interest, c
// Perform matrix multiplication in OpenBLAS (fortran call)
float alpha1 = 1.0;
float beta1 = 0.0;
char *nT = "N";
sgemm_(nT, nT, &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
sgemm_("N", "N", &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
// The above is a faster version of this, by calling the fortran version directly
//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);
@@ -585,13 +583,13 @@ void CEN_patch_expert::ResponseSparse(const cv::Mat_<float> &area_of_interest, c
void CEN_patch_expert::ResponseSparse_mirror(const cv::Mat_<float> &area_of_interest, cv::Mat_<float> &response, cv::Mat_<float>& mapMatrix, cv::Mat_<float>& im2col_prealloc)
{
int response_height = area_of_interest.rows - height + 1;
int response_width = area_of_interest.cols - width + 1;
int response_height = area_of_interest.rows - height_support + 1;
int response_width = area_of_interest.cols - width_support + 1;
cv::flip(area_of_interest, area_of_interest, 1);
// Extract im2col but in a sparse way and contrast normalize
im2colBiasSparseContrastNorm(area_of_interest, width, height, im2col_prealloc);
im2colBiasSparseContrastNorm(area_of_interest, width_support, height_support, im2col_prealloc);
response = im2col_prealloc.t();
@@ -610,8 +608,7 @@ void CEN_patch_expert::ResponseSparse_mirror(const cv::Mat_<float> &area_of_inte
// Perform matrix multiplication in OpenBLAS (fortran call)
float alpha1 = 1.0;
float beta1 = 0.0;
char *nT = "N";
sgemm_(nT, nT, &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
sgemm_("N", "N", &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
// The above is a faster version of this, by calling the fortran version directly
//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);
@@ -665,14 +662,14 @@ void CEN_patch_expert::ResponseSparse_mirror(const cv::Mat_<float> &area_of_inte
void CEN_patch_expert::ResponseSparse_mirror_joint(const cv::Mat_<float> &area_of_interest_left, const cv::Mat_<float> &area_of_interest_right, cv::Mat_<float> &response_left, cv::Mat_<float> &response_right, cv::Mat_<float>& mapMatrix, cv::Mat_<float>& im2col_prealloc_left, cv::Mat_<float>& im2col_prealloc_right)
{
int response_height = area_of_interest_left.rows - height + 1;
int response_width = area_of_interest_left.cols - width + 1;
int response_height = area_of_interest_left.rows - height_support + 1;
int response_width = area_of_interest_left.cols - width_support + 1;
cv::flip(area_of_interest_right, area_of_interest_right, 1);
// Extract im2col but in a sparse way and contrast normalize
im2colBiasSparseContrastNorm(area_of_interest_left, width, height, im2col_prealloc_left);
im2colBiasSparseContrastNorm(area_of_interest_right, width, height, im2col_prealloc_right);
im2colBiasSparseContrastNorm(area_of_interest_left, width_support, height_support, im2col_prealloc_left);
im2colBiasSparseContrastNorm(area_of_interest_right, width_support, height_support, im2col_prealloc_right);
cv::Mat_<float> response;
cv::vconcat(im2col_prealloc_left, im2col_prealloc_right, response);
@@ -695,8 +692,7 @@ void CEN_patch_expert::ResponseSparse_mirror_joint(const cv::Mat_<float> &area_o
// Perform matrix multiplication in OpenBLAS (fortran call)
float alpha1 = 1.0;
float beta1 = 0.0;
char *nT = "N";
sgemm_(nT, nT, &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
sgemm_("N", "N", &resp.cols, &weight.rows, &weight.cols, &alpha1, m1, &resp.cols, m2, &weight.cols, &beta1, m3, &resp.cols);
// The above is a faster version of this, by calling the fortran version directly
//cblas_sgemm(CblasColMajor, CblasNoTrans, CblasNoTrans, resp.cols, weight.rows, weight.cols, 1, m1, resp.cols, m2, weight.cols, 0.0, m3, resp.cols);