After using this convolution to replace the backbone network convolution, the parameter quantity decreases??

[yjqy12138]

After using this convolution to replace the backbone network convolution, the parameter quantity decreases??

[chaoli-ai]

Replacing vanilla convolution with ODConv will introduce more parameters due to the attention module. Please check your code carefully.

[yjqy12138]

I replaced the convolution in the official code of yolox, and found that the parameter quantity actually decreased