CodePudding user response:
33 DSST standard scale is very, very sensitive, easy to reduce the dimension number, even if you increase the corresponding step length, scale filter will be completely failed to keep pace with scale changes, may about this explanation is that the training scale filter using one dimensional samples, and there is no cyclic shift, which means that a training update only 33 samples, if reduce the number of samples, can lead to inadequate training, classifier understanding serious decline, not shift as there are so many translation filter samples, anyway, please don't try to dramatically reduce scale quantity, if you use scale filter 33 and 1.02 is very good,