Fig. 3
From: DeepSort: deep convolutional networks for sorting haploid maize seeds
Figure (a, b) shows the activations of all neurons in the convolutional layer 1 (each row corresponds to the activations that share a kernel) across images of 15 random diploid (a) and haploid (b) seeds (each column for a seed shown at the top row) from the test data set. Similar to figures (a, b), figures (c, d) shows the activations of neurons in the convolutional layer 2 across the same set of seeds. Kernels in the convolutional layers 1 and 2 perform various feature extractions and their complex compositions. For example, kernels 3 of first layer segments the seed from the background, and kernel 5 of the second layer provides discriminatory features (for other examples see text). Figure (e) shows visualizations of 16 kernels from the convolutional layer 1