WongKinYiu / yolov9

Implementation of paper - YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information
GNU General Public License v3.0
8.95k stars 1.42k forks source link

Hello, why is the pooling layer placed before the convolution layer in downsampling, while I've learned that it's usually placed after? What advantages does this approach offer? #580

Open magic524 opened 1 month ago

p-dot-max commented 1 month ago

There could be several reasons for this:

1.Reducing the spatial dimensions early on decreases the computational load in subsequent layers. 2.Pooling before convolution emphasizes the most significant features. 3.Noise suppression.