OpenCV Error: Bad Argument <Can not get new positive sample....

edit

The training of traincascade requires you to provide it with positive images (images having object) and negative images (images without the objects). The .vec file contains the information of all the images which are to be used by training network in binary format. Now when you are training the cascade, the numPos and numNeg indicate the number of samples which cascade should be using for training. So numPos and numNeg cannot be left undeclared else some default value for each gets assigned which are mostly 2000 and 1000 respectively which ultimately would give an error.

Also note that the numPos should always be less than the number of images used to generate vec file. This is because every time an image is falsely classified as being negative (i.e the image contains object but cascade couldn not locate it), unconsumed images are picked for training process. If the numPos is eqaul to the number of images used for vec file, it wont be left with any unconsumed images, leaving you with 'Can not get new positive sample' error.