yolov2layers, featurelayer and reorglayer

Just been working on yolov2Layers, and I've noticed that in both the help page on yolov2layers and the deep learning onramp courses, the featurelayer is relulayer after the reorglayer. Does the featurelayer always have to be after the reorglayer and what exactly is the point of the reorglayer?

Answer 1

[Omega]

Hi Hayden,

In YOLOv2 (You Only Look Once version 2) within MATLAB, the 'reorgLayer' and 'featureLayer' have specific roles to enhance object detection.

The 'reorgLayer' reshapes the feature map by merging high-resolution and low-resolution features, reducing spatial dimensions while increasing depth. This helps in detecting objects of various sizes.

The 'featureLayer', often a 'reluLayer', follows the 'reorgLayer' to introduce non-linearity, which is essential for learning complex patterns.

The reson for placing 'featureLayer' (e.g., 'reluLayer') after the 'reorgLayer' to apply activation functions to the reorganized features, enabling the network to learn more complex representations.

I hope it helps!