// Simple classifier that is 50% certain that the annotation is correct
Eigen::MatrixXf computeUnary( const VectorXs & lbl, int M ){
const float u_energy = -log( 1.0 / M );
const float n_energy = -log( (1.0 - GT_PROB) / (M-1) );
const float p_energy = -log( GT_PROB );
Eigen::MatrixXf r( M, lbl.rows() );
r.fill(u_energy);
//printf("%d %d %d \n",im[0],im[1],im[2]);
for( int k=0; k<lbl.rows(); k++ ){
// Set the energy
if (lbl[k]>=0){
r.col(k).fill( n_energy );
r(lbl[k],k) = p_energy;
}
}
return r;
}
Hamming::Hamming( const VectorXs & gt, float class_weight_pow ):gt_( gt ) {
int M=0,N=gt.rows();;
for( int i=0; i<N; i++ )
if( gt[i] >= M )
M = gt[i]+1;
VectorXf cnt = VectorXf::Zero( M );
for( int i=0; i<N; i++ )
if( gt[i] >= 0 )
cnt[gt[i]] += 1;
class_weight_ = cnt.array() / cnt.array().sum();
class_weight_ = class_weight_.array().pow( -class_weight_pow );
class_weight_ = class_weight_.array() / (cnt.array()*class_weight_.array()).sum();
}
