Skip to main content
. Author manuscript; available in PMC: 2018 Oct 1.
Published in final edited form as: IEEE Trans Image Process. 2017 Jun 28;26(10):4753–4764. doi: 10.1109/TIP.2017.2721106

Fig. 3.

Fig. 3

2D illustration of the fully convolutional network (FCN). Given an input 2D image, FCN first extracts multiple local image patches in a grid with the step size of np via the first max-pooling operation (with the kernel size of k × k). Here, we denote the orange and blue blocks as two p × p patches. The estimated displacements of these two grid-sampled patches are the two 1 × 1 × 3Nl elements in the outputs of FCN for one specific landmark.