shrec2015_.cpp

#include "SparseConvNet.h"
#include "SpatiallySparseDatasetSHREC2015.h"

int epoch=0;
int cudaDevice=-1;
int batchSize=10;

int nFeaturesPerLevel(int i) {
  return 32*(i+1); //This can be increased
}

class DeepPOFMP : public SparseConvNet {
public:
  DeepPOFMP(int dimension, int nInputFeatures, int nClasses, int cudaDevice,int nTop=1) : SparseConvNet(dimension, nInputFeatures, nClasses, cudaDevice) {
    const float fmpShrink=powf(2,0.6666); //Fractional Max-Pooling ratio.
    int l=7; //Number of levels of FMP
    for (int i=0;i<l;i++) {
      addLeNetLayerPOFMP(32*(i+1),2,1,2,fmpShrink,VLEAKYRELU);
    }
    addLeNetLayerMP(32*(l+1),2,1,1,1,VLEAKYRELU);
    addLeNetLayerMP(32*(l+2),1,1,1,1,VLEAKYRELU);
    addSoftmaxLayer();
  }
};

int main(int lenArgs, char *args[]) {
  std::string baseName="weights/SHREC2015_";
  int fold=0;
  if (lenArgs>1)
    fold=atoi(args[1]);
  std::cout << "Fold: " << fold << std::endl;
  SpatiallySparseDataset trainSet=SHREC2015TrainSet(32,6,fold);
  trainSet.summary();
  trainSet.repeatSamples(10);
  SpatiallySparseDataset testSet=SHREC2015TestSet(32,6,fold);
  testSet.summary();

  DeepPOFMP cnn(3,trainSet.nFeatures,trainSet.nClasses,cudaDevice);
  if (epoch>0)
    cnn.loadWeights(baseName,epoch);
  for (epoch++;epoch<=200;epoch++) {
    std::cout <<"epoch: " << epoch << std::flush;
    cnn.processDataset(trainSet, batchSize,0.003*exp(-0.05/2 * epoch));
    if (epoch%20==0) {
      cnn.saveWeights(baseName,epoch);
      cnn.processDatasetRepeatTest(testSet,  batchSize, 3);
    }
  }
}