The classification subnet applies four 3 × 3 convolutionlayers each with 256 filters, followed by a 3×3 convolutionlayer with KA filters where K means the number of classesand A means the number of anchors per location.
The classification subnet applies four 3 × 3 convolution<br>layers each with 256 filters, followed by a 3×3 convolution<br>layer with KA filters where K means the number of classes<br>and A means the number of anchors per location.