Re-train a model..

[hector246288]

Hi there, I have downloaded the AFLW database and started to train my own model. The numbers of positive and negative samples are more than 20,000. However, the outcome doesn't look well. Here are my parameters：-objSize 20 -numPos 20000 -negRatio 1 -maxTreeLevel 8 -minDR 1 -maxFAR 1e-5 -maxNumStages 200 -ifFlip 1

Could you help me figure out what the problem was?

[EvergreenHZ]

1. increase negRatio (say -negRatio 1.5) and decrease maxFAR (say -maxFAR 1e-8)
2. find a good threshold score in the test case (say float score = 10 instead of -1, see bellow) `int main(int argc, char* argv[]) {

printf("************* [TEST] Npd:detect test... *************\n");

npd::npddetect npd;
npd.load(argv[1]);
cv::Mat img = cv::imread(argv[2], 0);
string savepath = "1.jpg";
float score = -1;  # change -1 to other number`