Architecture of Deep Convolutional Encoder-Decoder Networks for Building Footprint Semantic Segmentation

Figure 12

Visual comparisons. White = TP (the predicted result is a building, and the ground truth is also a building), black = TN (the ground truth is a non-building, and the predicted result is a non-building), yellow = FN (the ground truth is a building, and the predicted result is a non-building), and red = FP (the ground truth is a non-building, but the predicted result is a building). (a) Test image, (b) ground truth, (c) PSPNet-VGG16, (d) LinkNet-VGG16, and (e) U-Net-VGG16.

(a)

(b)

(c)

(d)

(e)