regression = self.regression(features_fpn[1]) * 100

[SherlockHolmes221]

why plus 100 here? thanks

[wzr0108]

I guess it’s because anchor_points are the coordinates of the original image, and the network output is too small and needs to be enlarged.