// Training SVM with feature vector X and label Y.
// Each row of X is a feature vector, with corresponding label in Y.
// Return a CV_32F weight Mat
Mat Objectness::trainSVM(CMat &X1f, const vecI &Y, int sT, double C, double bias, double eps)
{
// Set SVM parameters
parameter param; {
param.solver_type = sT; // L2R_L2LOSS_SVC_DUAL;
param.C = C;
param.eps = eps; // see setting below
param.p = 0.1;
param.nr_weight = 0;
param.weight_label = NULL;
param.weight = NULL;
set_print_string_function(print_null);
CV_Assert(X1f.rows == Y.size() && X1f.type() == CV_32F);
}
// Initialize a problem
feature_node *x_space = NULL;
problem prob;{
prob.l = X1f.rows;
prob.bias = bias;
prob.y = Malloc(double, prob.l);
prob.x = Malloc(feature_node*, prob.l);
const int DIM_FEA = X1f.cols;
prob.n = DIM_FEA + (bias >= 0 ? 1 : 0);
x_space = Malloc(feature_node, (prob.n + 1) * prob.l);
int j = 0;
for (int i = 0; i < prob.l; i++){
prob.y[i] = Y[i];
prob.x[i] = &x_space[j];
const float* xData = X1f.ptr<float>(i);
for (int k = 0; k < DIM_FEA; k++){
x_space[j].index = k + 1;
x_space[j++].value = xData[k];
}
if (bias >= 0){
x_space[j].index = prob.n;
x_space[j++].value = bias;
}
x_space[j++].index = -1;
}
CV_Assert(j == (prob.n + 1) * prob.l);
}
// Training SVM for current problem
const char* error_msg = check_parameter(&prob, ¶m);
if(error_msg){
fprintf(stderr,"ERROR: %s\n",error_msg);
exit(1);
}
model *svmModel = train(&prob, ¶m);
Mat wMat(1, prob.n, CV_64F, svmModel->w);
wMat.convertTo(wMat, CV_32F);
free_and_destroy_model(&svmModel);
destroy_param(¶m);
free(prob.y);
free(prob.x);
free(x_space);
return wMat;
}
bool CIntersectionAssistanceUnit::GetHighestScoringLastKnownGoodPosition( const QuatT& baseOrientation, QuatT& outQuat ) const
{
bool bFlippedIsBest = false;
if(!m_lastKnownGoodPositions.empty())
{
// Higher is better
float fBestScore = 0.0f;
int bestIndex = -1;
Vec3 vBaseUpDir = baseOrientation.q.GetColumn2().GetNormalized();
for(uint8 i = 0; i < m_lastKnownGoodPositions.size(); ++i)
{
const QuatT& qLastKnownGood = m_lastKnownGoodPositions[i];
if(IsPositionWithinAcceptedLimits(qLastKnownGood.t, baseOrientation.t, kDistanceTolerance))
{
// Generate [0.0f,1.0f] score for distance
const Vec3 distVec = (qLastKnownGood.t - baseOrientation.t);
const float length = max(distVec.GetLengthFast(),0.0001f);
const float distanceScore = max(1.0f - (length * kInverseDistanceTolerance) * kDistanceWeight, 0.0f);
Vec3 vUpDir = qLastKnownGood.q.GetColumn2();
const float regularOrientationScore = vBaseUpDir.Dot(vUpDir);
const float flippedOrientationScore = vBaseUpDir.Dot(-vUpDir);
float orientationScore = max(regularOrientationScore, flippedOrientationScore);
orientationScore *= kOrientationWeight;
const float fCandidateScore = distanceScore + orientationScore;
#ifndef _RELEASE
if(g_pGameCVars->pl_pickAndThrow.intersectionAssistDebugEnabled == 2)
{
CryWatch("[INDEX(%d)] : D[%.3f] O[%.3f] T[%.3f] (%s)", i, distanceScore, orientationScore, fCandidateScore, flippedOrientationScore > regularOrientationScore ? "*F*" : "R");
}
#endif //#ifndef _RELEASE
if(fCandidateScore > fBestScore)
{
bestIndex = i;
fBestScore = fCandidateScore;
bFlippedIsBest = (flippedOrientationScore > regularOrientationScore);
}
}
}
if(bestIndex >= 0)
{
outQuat = m_lastKnownGoodPositions[bestIndex];
if(bFlippedIsBest)
{
Matrix34 wMat(outQuat);
Vec3 vFlippedUpDir = -outQuat.q.GetColumn2().GetNormalized();
Vec3 vForwardDir = outQuat.q.GetColumn1().GetNormalized();
Vec3 vSideDir = -outQuat.q.GetColumn0().GetNormalized();
Matrix34 wFlippedMat;
wFlippedMat = Matrix34::CreateFromVectors(vSideDir, vForwardDir, vFlippedUpDir, wMat.GetTranslation());
outQuat = QuatT(wFlippedMat);
// Adjust pos (rotating around OOBB centre effectively)
const IEntity* pSubjectEntity = gEnv->pEntitySystem->GetEntity(m_subjectEntityId);
if(pSubjectEntity)
{
AABB entAABB;
OBB entOBB;
pSubjectEntity->GetLocalBounds(entAABB);
entOBB.SetOBBfromAABB(Quat(IDENTITY), entAABB);
Vec3 Centre = wMat.TransformPoint(entOBB.c);
Vec3 toCentre = Centre - outQuat.t;
outQuat.t += (toCentre * 2.0f);
}
}
#ifndef _RELEASE
if(g_pGameCVars->pl_pickAndThrow.intersectionAssistDebugEnabled == 2)
{
m_currentBestIndex = bestIndex;
CryWatch("[BEST INDEX] : %d", bestIndex);
}
#endif // ifndef _RELEASE
return true;
}
}
#ifndef _RELEASE
m_currentBestIndex = -1;
#endif // ifndef _RELEASE
return false;
}
void Objectness::trainStateII(int numPerSz)
{
loadTrainedModel();
const int NUM_TRAIN = _voc.trainNum;
vector<vecI> SZ(NUM_TRAIN), Y(NUM_TRAIN);
vector<vecF> VAL(NUM_TRAIN);
#pragma omp parallel for
for (int i = 0; i < _voc.trainNum; i++) {
const vector<Vec4i> &bbgts = _voc.gtTrainBoxes[i];
ValStructVec<float, Vec4i> valBoxes;
vecI &sz = SZ[i], &y = Y[i];
vecF &val = VAL[i];
CStr imgPath = format(_S(_voc.imgPathW), _S(_voc.trainSet[i]));
predictBBoxSI(imread(imgPath), valBoxes, sz, numPerSz, false);
const int num = valBoxes.size();
CV_Assert(sz.size() == num);
y.resize(num), val.resize(num);
for (int j = 0; j < num; j++) {
Vec4i bb = valBoxes[j];
val[j] = valBoxes(j);
y[j] = maxIntUnion(bb, bbgts) >= 0.5 ? 1 : -1;
}
}
const int NUM_SZ = _svmSzIdxs.size();
const int maxTrainNum = 100000;
vector<vecM> rXP(NUM_SZ), rXN(NUM_SZ);
for (int r = 0; r < NUM_SZ; r++) {
rXP[r].reserve(maxTrainNum);
rXN[r].reserve(1000000);
}
for (int i = 0; i < NUM_TRAIN; i++) {
const vecI &sz = SZ[i], &y = Y[i];
vecF &val = VAL[i];
int num = sz.size();
for (int j = 0; j < num; j++) {
int r = sz[j];
CV_Assert(r >= 0 && r < NUM_SZ);
if (y[j] == 1)
rXP[r].push_back(Mat(1, 1, CV_32F, &val[j]));
else
rXN[r].push_back(Mat(1, 1, CV_32F, &val[j]));
}
}
Mat wMat(NUM_SZ, 2, CV_32F);
for (int i = 0; i < NUM_SZ; i++) {
const vecM &xP = rXP[i], &xN = rXN[i];
if (xP.size() < 10 || xN.size() < 10)
printf("Warning %s:%d not enough training sample for r[%d] = %d. P = %d, N = %d\n", __FILE__, __LINE__, i, _svmSzIdxs[i], xP.size(), xN.size());
for (size_t k = 0; k < xP.size(); k++)
CV_Assert(xP[k].size() == Size(1, 1) && xP[k].type() == CV_32F);
Mat wr = trainSVM(xP, xN, L1R_L2LOSS_SVC, 100, 1);
CV_Assert(wr.size() == Size(2, 1));
wr.copyTo(wMat.row(i));
}
matWrite(_modelName + ".wS2", wMat);
_svmReW1f = wMat;
}
