Skip to main content
. 2022 Mar 1;12:3383. doi: 10.1038/s41598-022-07217-0

Figure 6.

Figure 6

The domain distribution is illustrated. The scatter plots are t-SNE embeddings of 745 picture tiles chosen at random, with representative patches displayed on the left and framed in corresponding colors. The feature map acquired from the activation of the FCN network's last convolutional layer is used to represent each image tile. The results of two experiments are compared: (a) representation learned by a baseline FCN model and (b) domain agnostic training.