Figure 2 | Scientific Reports

Figure 2

From: Orthogonal Representations of Object Shape and Category in Deep Convolutional Neural Networks and Human Visual Cortex

Figure 2

Correlation between layers in CNNs and shape (orange) versus category (blue) in Set A (top row) and B (bottom row). The horizontal axis indicates network depth and the vertical axis indicates correlation (Spearman’s ρ). For GoogLeNet and ResNet architectures, the correlations shown are for 3 × 3 convolutional operations, while other parallel operations (projections and convolutions of different sizes) are omitted. Dashed line indicates significance threshold of p < 0.05, which was calculated by randomly permuting the RDM labels and then calculating dissimilarity relationships 1000 times. Grey shading indicates fully-connected layers. Results for Set B CaffeNet, VGG-19 and GoogLeNet were previously reported in Kubilius et al.18.

Back to article page