Feature maps extracted from different layers of a convolutional neural network possess varying-size receptive fields with respect to the input image. This hierarchical structure can be leveraged for multiscale object detection by designing a network where feature maps closer to the output layer, which have wider receptive fields, are responsible for detecting larger objects, while shallower feature maps detect smaller objects.

Multiscale Object Detection via Receptive Fields

In a feature map with $$c$$ channels and shape $$h 	imes w$$, the $$c$$ units at any single spatial position share the exact same receptive field on the input image. These $$c$$ units are transformed to predict the classes and offsets of the $$a$$ anchor boxes that are generated using that specific spatial position. Consequently, the model leverages the visual information contained within a specific receptive field to predict the properties of anchor boxes located spatially close to it.

Claude

At a specific scale represented by a feature map of shape $$h 	imes w$$, a convolutional neural network generates $$hw$$ sets of anchor boxes, with each set containing $$a$$ anchor boxes sharing the same center. Once each anchor box is labeled with a ground-truth class and offset, the object detection model must predict these classes and offsets for all $$hw$$ sets of anchor boxes across the input image.

Learn Before

Related

Learn After