search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.68k stars 7.96k forks source link

convolution subsampling #4592

Open leiyaohui opened 4 years ago

leiyaohui commented 4 years ago

Hello, I want to know how to realize the following two kinds of convolution subsampling rounding problems? Can tell these two kinds where to modify, and then recompile. 4 conv 64 3 x 3 / 1 117 x 117 x 32 -> 117 x 117 x 64 0.505 BF 5 max 2 x 2 / 2 117 x 117 x 64 -> 59 x 59 x 64 0.001 BF

4 conv 64 3 x 3 / 1 117 x 117 x 32 -> 117 x 117 x 64 0.505 BF 5 max 2 x 2 / 2 117 x 117 x 64 -> 58 x 58 x 64 0.001 BF I look forward to your reply！

AlexeyAB commented 4 years ago

use padding=0

[maxpool]
padding=0
size=2
stride=2