Pp-yoloe details

        PP-YOLOE There are a series of models , namely s/m/l/x, Can pass width multiplier and depth multiplier To configure .PP-YOLOE Avoid using things like deformable convolution perhaps matrix nms Special operators like , So that it can be easily deployed on a variety of hardware .

The design mechanism of the detector includes ：

• Anchor free Mechanism

• Extensible backbone and neck, from CSPRepResStage(CSPNet+RMNet) constitute

• Use Varifocal Loss(VFL) and Distribution focal loss(DFL) Head mechanism ET-head(Efficient Task-aligned head)

• Dynamic label assignment algorithm Task Alignment Learning（TAL）

• SiLU Activation function

Model architecture ：

 PP-YOLOE Of Backbone Mainly used RepVGG Modules and CSP The model idea is right ResNet And retrograde improvements , It also uses SiLU Activation function 、Effitive SE Attention Equal module .

RepVGG

RepVGG, This network is in VGG Improve on the basis of , The main ideas include ：

<