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

Table 1

Comparison metrics of the models tested in this study.

Model	Encoder	IOU	F1-score	Precision	Recall
PSPNet	VGG16	0.7784	0.8743	0.8405	0.911
	VGG19	0.7772	0.8735	0.8426	0.9068
	ResNet50	0.7176	0.8452	0.8261	0.8653
	DenseNet169	0.6669	0.7946	0.7762	0.8139
	Xception	0.6528	0.7844	0.7566	0.8144
LinkNet	VGG16	0.8179	0.8992	0.8585	0.9439
	VGG19	0.8193	0.9001	0.8568	0.9481
	ResNet50	0.82	0.9003	0.8671	0.9363
	DenseNet169	0.8238	0.9027	0.8808	0.9256
	Xception	0.8127	0.8959	0.8715	0.9216
FPN	VGG16	0.6952	0.7681	0.7403	0.7981
	VGG19	0.6812	0.7528	0.7252	0.7826
	ResNet50	0.6656	0.7481	0.7255	0.7721
	DenseNet169	0.6632	0.7476	0.7288	0.7673
	Xception	0.6471	0.7336	0.7120	0.7567
U-Net	VGG16	0.8302	0.9067	0.8846	0.9298
	VGG19	0.8296	0.9064	0.8827	0.9314
	ResNet50	0.8233	0.9023	0.8741	0.9324
	DenseNet169	0.826	0.9041	0.8788	0.931
	Xception	0.821	0.901	0.8782	0.925