True Orthoimage Generation Using Airborne LiDAR Data with Generative Adversarial Network-Based Deep Learning Model

Table 3

Evaluation of training performance for Potsdam unseen datasets.

Experiment	Training data type	Tile size (pixel)	Overlap (%)	FID	SSIM
1	DSM		50	451.39	0.375
2	Color label		50	510.94	0.352
3	B/W label		50	515.57	0.363
4	DSM+B/W label		50	519.37	0.384
5	DSM		60	515.69	0.377
6	DSM		50	498.72	0.368
7	DSM		70	442.75	0.377
8	DSM		50	526.68	0.346
9	DSM		80	492.61	0.370
10	DSM+B/W label		80	373.03	0.387