// compute gradient magnitude and orientation at each location (uses sse)
void gradMag( float *I, float *M, float *O, int h, int w, int d ) {
int x, y, y1, c, h4, s; float *Gx, *Gy, *M2; __m128 *_Gx, *_Gy, *_M2, _m;
float *acost = acosTable(), acMult=25000/2.02f;
// allocate memory for storing one column of output (padded so h4%4==0)
h4=(h%4==0) ? h : h-(h%4)+4; s=d*h4*sizeof(float);
M2=(float*) alMalloc(s,16); _M2=(__m128*) M2;
Gx=(float*) alMalloc(s,16); _Gx=(__m128*) Gx;
Gy=(float*) alMalloc(s,16); _Gy=(__m128*) Gy;
// compute gradient magnitude and orientation for each column
for( x=0; x<w; x++ ) {
// compute gradients (Gx, Gy) and squared magnitude (M2) for each channel
for( c=0; c<d; c++ ) grad1( I+x*h+c*w*h, Gx+c*h4, Gy+c*h4, h, w, x );
for( y=0; y<d*h4/4; y++ ) _M2[y]=ADD(MUL(_Gx[y],_Gx[y]),MUL(_Gy[y],_Gy[y]));
// store gradients with maximum response in the first channel
for(c=1; c<d; c++) {
for( y=0; y<h4/4; y++ ) {
y1=h4/4*c+y; _m = CMPGT( _M2[y1], _M2[y] );
_M2[y] = OR( AND(_m,_M2[y1]), ANDNOT(_m,_M2[y]) );
_Gx[y] = OR( AND(_m,_Gx[y1]), ANDNOT(_m,_Gx[y]) );
_Gy[y] = OR( AND(_m,_Gy[y1]), ANDNOT(_m,_Gy[y]) );
}
}
// compute gradient mangitude (M) and normalize Gx
for( y=0; y<h4/4; y++ ) {
_m = MIN( RCPSQRT(_M2[y]), SET(1e10f) );
_M2[y] = RCP(_m);
_Gx[y] = MUL( MUL(_Gx[y],_m), SET(acMult) );
_Gx[y] = XOR( _Gx[y], AND(_Gy[y], SET(-0.f)) );
};
memcpy( M+x*h, M2, h*sizeof(float) );
// compute and store gradient orientation (O) via table lookup
if(O!=0) for( y=0; y<h; y++ ) O[x*h+y] = acost[(int)Gx[y]];
}
alFree(Gx); alFree(Gy); alFree(M2);
}
// compute gradient magnitude and orientation at each location (uses sse)
void gradMag( float *I, float *M, float *O, int h, int w, int d, bool full ) {
int x, y, y1, c, h4, s;
float *Gx, *Gy, *M2;
__m128 *_Gx, *_Gy, *_M2, _m;
float *acost = acosTable(), acMult = 10000.0f;
// allocate memory for storing one column of output (padded so h4%4==0)
h4 = (h % 4 == 0) ? h : h - (h % 4) + 4;
s = d * h4 * sizeof(float);
M2 = (float*) alMalloc(s, 16);
_M2 = (__m128*) M2;
Gx = (float*) alMalloc(s, 16);
_Gx = (__m128*) Gx;
Gy = (float*) alMalloc(s, 16);
_Gy = (__m128*) Gy;
// compute gradient magnitude and orientation for each column
for ( x = 0; x < w; x++ ) {
// compute gradients (Gx, Gy) with maximum squared magnitude (M2)
for (c = 0; c < d; c++) {
grad1( I + x * h + c * w * h, Gx + c * h4, Gy + c * h4, h, w, x );
for ( y = 0; y < h4 / 4; y++ ) {
y1 = h4 / 4 * c + y;
_M2[y1] = ADD(MUL(_Gx[y1], _Gx[y1]), MUL(_Gy[y1], _Gy[y1]));
if ( c == 0 ) { continue; }
_m = CMPGT( _M2[y1], _M2[y] );
_M2[y] = OR( AND(_m, _M2[y1]), ANDNOT(_m, _M2[y]) );
_Gx[y] = OR( AND(_m, _Gx[y1]), ANDNOT(_m, _Gx[y]) );
_Gy[y] = OR( AND(_m, _Gy[y1]), ANDNOT(_m, _Gy[y]) );
}
}
// compute gradient mangitude (M) and normalize Gx
for ( y = 0; y < h4 / 4; y++ ) {
_m = MINsse( RCPSQRT(_M2[y]), SET(1e10f) );
_M2[y] = RCP(_m);
if (O) { _Gx[y] = MUL( MUL(_Gx[y], _m), SET(acMult) ); }
if (O) { _Gx[y] = XOR( _Gx[y], AND(_Gy[y], SET(-0.f)) ); }
};
memcpy( M + x * h, M2, h * sizeof(float) );
// compute and store gradient orientation (O) via table lookup
if ( O != 0 ) for ( y = 0; y < h; y++ ) { O[x * h + y] = acost[(int)Gx[y]]; }
if ( O != 0 && full ) {
y1 = ((~size_t(O + x * h) + 1) & 15) / 4;
y = 0;
for ( ; y < y1; y++ ) { O[y + x * h] += (Gy[y] < 0) * PI; }
for ( ; y < h - 4; y += 4 ) STRu( O[y + x * h],
ADD( LDu(O[y + x * h]), AND(CMPLT(LDu(Gy[y]), SET(0.f)), SET(PI)) ) );
for ( ; y < h; y++ ) { O[y + x * h] += (Gy[y] < 0) * PI; }
}
}
alFree(Gx);
alFree(Gy);
alFree(M2);
}
void
pcl::people::HOG::gradMag( float *I, int h, int w, int d, float *M, float *O ) const
{
#if defined(__SSE2__)
int x, y, y1, c, h4, s; float *Gx, *Gy, *M2; __m128 *_Gx, *_Gy, *_M2, _m;
float *acost = acosTable(), acMult=25000/2.02f;
// allocate memory for storing one column of output (padded so h4%4==0)
h4=(h%4==0) ? h : h-(h%4)+4; s=d*h4*sizeof(float);
M2=(float*) alMalloc(s,16);
_M2=(__m128*) M2;
Gx=(float*) alMalloc(s,16); _Gx=(__m128*) Gx;
Gy=(float*) alMalloc(s,16); _Gy=(__m128*) Gy;
// compute gradient magnitude and orientation for each column
for( x=0; x<w; x++ ) {
// compute gradients (Gx, Gy) and squared magnitude (M2) for each channel
for( c=0; c<d; c++ ) grad1( I+x*h+c*w*h, Gx+c*h4, Gy+c*h4, h, w, x );
for( y=0; y<d*h4/4; y++ ) _M2[y]=ADD(MUL(_Gx[y],_Gx[y]),MUL(_Gy[y],_Gy[y]));
// store gradients with maximum response in the first channel
for(c=1; c<d; c++) {
for( y=0; y<h4/4; y++ ) {
y1=h4/4*c+y; _m = CMPGT( _M2[y1], _M2[y] );
_M2[y] = OR( AND(_m,_M2[y1]), ANDNOT(_m,_M2[y]) );
_Gx[y] = OR( AND(_m,_Gx[y1]), ANDNOT(_m,_Gx[y]) );
_Gy[y] = OR( AND(_m,_Gy[y1]), ANDNOT(_m,_Gy[y]) );
}
}
// compute gradient magnitude (M) and normalize Gx
for( y=0; y<h4/4; y++ ) {
_m = MIN( RCPSQRT(_M2[y]), SET(1e10f) );
_M2[y] = RCP(_m);
_Gx[y] = MUL( MUL(_Gx[y],_m), SET(acMult) );
_Gx[y] = XOR( _Gx[y], AND(_Gy[y], SET(-0.f)) );
};
memcpy( M+x*h, M2, h*sizeof(float) );
// compute and store gradient orientation (O) via table lookup
if(O!=0) for( y=0; y<h; y++ ) O[x*h+y] = acost[(int)Gx[y]];
}
alFree(Gx); alFree(Gy); alFree(M2);
#else
int x, y, y1, c, h4, s; float *Gx, *Gy, *M2;
float *acost = acosTable(), acMult=25000/2.02f;
// allocate memory for storing one column of output (padded so h4%4==0)
h4=(h%4==0) ? h : h-(h%4)+4; s=d*h4*sizeof(float);
M2=(float*) alMalloc(s,16);
Gx=(float*) alMalloc(s,16);
Gy=(float*) alMalloc(s,16);
float m;
// compute gradient magnitude and orientation for each column
for( x=0; x<w; x++ ) {
// compute gradients (Gx, Gy) and squared magnitude (M2) for each channel
for( c=0; c<d; c++ ) grad1( I+x*h+c*w*h, Gx+c*h4, Gy+c*h4, h, w, x );
for( y=0; y<d*h4; y++ )
{
M2[y] = Gx[y] * Gx[y] + Gy[y] * Gy[y];
}
// store gradients with maximum response in the first channel
for(c=1; c<d; c++)
{
for( y=0; y<h4/4; y++ )
{
y1=h4/4*c+y;
for (int ii = 0; ii < 4; ++ii)
{
if (M2[y1 * 4 + ii] > M2[y * 4 + ii])
{
M2[y * 4 + ii] = M2[y1 * 4 + ii];
Gx[y * 4 + ii] = Gx[y1 * 4 + ii];
Gy[y * 4 + ii] = Gy[y1 * 4 + ii];
}
}
}
}
// compute gradient magnitude (M) and normalize Gx
for( y=0; y<h4; y++ )
{
m = 1.0f/sqrtf(M2[y]);
m = m < 1e10f ? m : 1e10f;
M2[y] = 1.0f / m;
Gx[y] = ((Gx[y] * m) * acMult);
if (Gy[y] < 0)
Gx[y] = -Gx[y];
}
memcpy( M+x*h, M2, h*sizeof(float) );
// compute and store gradient orientation (O) via table lookup
if(O!=0) for( y=0; y<h; y++ ) O[x*h+y] = acost[(int)Gx[y]];
}
alFree(Gx); alFree(Gy); alFree(M2);
#endif
}