Fig. 1: Schematic illustration of image similarity computational structure.
From: Symmetry-aware recursive image similarity exploration for materials microscopy
a VGG-16 architecture pretrained on ImageNet database. The gray area represents the pretrained VGG-16 model. FC3 was determined by training an autoencoder to reduce the dimensionality of the feature vector from 4096 to 512. The gear represents that this is the only layer optimized in the VGG-16 model. b ResNet-34 architecture with symmetry awareness by training it to classify wallpaper group symmetries.
