1

I'm developing an android app to detect object like chairs and doors etc. For that I am using cascade classifier to train data.I have collected 1000 of positive sample for chair of various type and around 700 of negative sample contain image without chairs. But I am getting error "Required leaf false alarm rate achieved". Then i research a bit and change my way of training as explain in one of the stack overflow answer. But still i am getting same error.And the training is not passing beyond stage 1.

Where am i going wrong? Here is my training command. enter image description here

Kamal Kumar Majhi
  • 51
  • 1
  • 12
  • post your training command. If the false alarm rate is already reached so early, there might be some bias in your data which is some kind of artificial difference between you positive and negative images. – Micka Mar 31 '17 at 05:33
  • increase the number of stages or decrease the maxFalseAlarmRate. 0.5^6 = 0.015625 which isn't so great (and can often be achieved in the first stage). – Micka Mar 31 '17 at 06:16
  • @Micka I have already tried to increase the stages by 15-16 but the training stages stuck after one day of training . Is it normal to leave the training for weeks or should i try different approach? And also i read from a source that detection of chair is very tough job.So what should i do? – Kamal Kumar Majhi Mar 31 '17 at 06:46
  • collevt more negative samples without increasing the numNeg parameter if the process gets stucked at finding negative samples. Bur yes, it is normal that this kind of training needs a lot of time. – Micka Mar 31 '17 at 06:48
  • Only 700 negative samples sounds like not enough by far. Probably the classifiers are training something else. – NoDataDumpNoContribution Mar 31 '17 at 13:09

0 Answers0