Fig 1.
Pipeline of PAM.
Fig 2.
Pipeline of CAM.
Fig 3.
Overall framework.
Fig 4.
Pipeline of SPAM.
Fig 5.
Pipeline of SCAM.
Fig 6.
Architecture of MLFD.
Fig 7.
Illustration of Potsdam dataset.
Fig 8.
Illustration of DeepGlobe dataset.
Table 1.
Hyper-parameter settings.
Table 2.
Data properties and partitions.
Table 3.
Comparative methods.
Table 4.
The results of the Potsdam test set with class-wise performance in form of IoU/OA and overall performance with mIoU and OA, where bold indicates the best.
Fig 9.
Visual inspections of random samples from the Potsdam test set.
Table 5.
The results of the DeepGlobe test set with class-wise performance in form of IoU/OA and overall performance with mIoU and OA, where bold indicates the best.
Fig 10.
Visual inspections of random samples from the DeepGlobe test set.
Table 6.
Accuracy comparisons in form of mIoU/OA on test set.
Fig 11.
Training loss on Potsdam dataset.
Fig 12.
Training mIoU on Potsdam dataset.
Fig 13.
Training loss on DeepGlobe dataset.
Fig 14.
Training mIoU on DeepGlobe dataset.
Table 7.
Efficiency comparisons.
Table 8.
Parameter comparisons.
Table 9.
Results of different decoders in form of mIoU/OA.