view article
Figure 6
Three examples of CNN representations at the first five convolutional layers of AlexNet. The network detects edges from pixels in the first layer, then uses those edges to detect shapes in the next layer, and then uses that result to infer complex shapes and objects in later layers. Thus, despite growing fuzziness, convolutional layers continue to maintain photographically accurate representations of input images. (Best viewed in colour and enlarged.)

Journal logoJOURNAL OF
APPLIED
CRYSTALLOGRAPHY
ISSN: 1600-5767
Volume 56| Part 5| October 2023| Pages 1494-1504
https://doi.org/10.1107/S1600576723007446
Follow J. Appl. Cryst.
Sign up for e-alerts
E-alerts
Follow J. Appl. Cryst. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS