
The ability of a CapsNet to recognize image attributes depends on the characteristics of the capsules. When recognizing the image, CapsNets can judge that the image is not a face. proposed a CNN with dynamic routing algorithm 3, 4.Ĭapsule networks (CapsNets) are effective at recognizing various attributes of specific entities in the image, including pose (position, size, direction), deformation, speed, reflectivity, hue, texture, and so on. In order to solve the problem of the traditional CNN being too coarse for image understanding, Hinton et al. For example, when processing an image with a nose, eyes, mouth, and other facial features but not a face, the CNN will stubbornly think that it is a face 1, 2.

When the recognition changes the image of position feature, the result of CNNs is not good. While local characteristics are improved, other internal information, such as position and attitude, can be lost.

A pooling operation can be regarded as a routing application. In CNNs, after the convolution operation of each layer is completed, a pooling operation is carried out.

In neural networks, routing is the process of transferring information from one layer to another. Routing refers to the process that determines the network scope of the end-to-end path of the packet from the source to the destination. However, CNNs are not perfect, and their ability to deal with the spatial relationships of image entities is inadequate. Convolutional neural networks (CNNs) are widely used in computer vision because of their great success in target recognition and classification.
