Skip to main content
. 2017 Dec 5;7:16991. doi: 10.1038/s41598-017-17299-w

Figure 3.

Figure 3

(a) Classification accuracies for the rotation invariance of the R3DVS descriptors. The horizontal axis indicates number of the rotated replicas that are fed into the CNNs. The vertical axis indicates classification accuracy averaged over 20 test results for randomly-chosen 50 samples of the oxides. (b) and (c) are t-SNE feature visualisations projected from the last Leaky ReLU layers embedded in the convolutional layers with 10 and 30 rotated replicas, respectively. The types of the marks indicate the 50 targets of the oxides to be classified.