Open dienthaipham103 opened 3 years ago
@dienthaipham103 It would longer time in training and furthermore during testing, you will have a huge number of overlapping boxes that would be difficult to suppress! I usually use default values given in the code, unless I a long object with a bigger ratio, only then i change ratios. For scales, it depends whether you are trying to predict smaller or larger objects only then you can choose a range for that
I got it. Thank you so much for your answer!
Does increasing the number of scales and ratios improve the performance of RPN model?