An Empirical Study of Remote Sensing Pretraining

Benchmark Model Rank Results
semantic-segmentation-on-isprs-potsdamIMP-ViTAEv2-S-UperNet#5Overall Accuracy: 91.6
semantic-segmentation-on-isprs-potsdamRSP-ViTAEv2-S-UperNet#10Overall Accuracy: 91.21
semantic-segmentation-on-isprs-potsdamRSP-Swin-T-UperNet#12Overall Accuracy: 90.78
semantic-segmentation-on-isprs-potsdamRSP-ResNet-50-UperNet#14Overall Accuracy: 90.61
change-detection-for-remote-sensing-images-on-cdd-season-varyingIMP-ViTAEv2-S-BIT#8F1-Score: 0.9702
change-detection-for-remote-sensing-images-on-cdd-season-varyingRSP-ViTAEv2-S-BIT#9F1-Score: 0.9681
change-detection-for-remote-sensing-images-on-cdd-season-varyingRSP-ResNet-50-BIT#11F1-Score: 0.96
change-detection-for-remote-sensing-images-on-cdd-season-varyingRSP-Swin-T-BIT#13F1-Score: 0.9521
object-detection-in-aerial-images-on-hrsc2016RSP-ViTAEv2-S-FPN-ORCN#9mAP-07: 90.4
object-detection-in-aerial-images-on-hrsc2016IMP-ViTAEv2-S-FPN-ORCN#10mAP-07: 90.4
object-detection-in-aerial-images-on-hrsc2016RSP-ResNet-50-FPN-ORCN#11mAP-07: 90.3
object-detection-in-aerial-images-on-hrsc2016RSP-Swin-T-FPN-ORCN#12mAP-07: 90.0
semantic-segmentation-on-isaidIMP-ViTAEv2-S-UperNet#12mIoU: 65.3
semantic-segmentation-on-isaidRSP-ViTAEv2-S-UperNet#15mIoU: 64.3
semantic-segmentation-on-isaidRSP-Swin-T-UperNet#16mIoU: 64.1
semantic-segmentation-on-isaidRSP-ResNet-50-UperNet#19mIoU: 61.6
building-change-detection-for-remote-sensing-images-on-levir-cdIMP-ViTAEv2-S-BIT#15F1: 91.26
building-change-detection-for-remote-sensing-images-on-levir-cdRSP-ViTAEv2-S-BIT#20F1: 90.93IoU: 84.95
building-change-detection-for-remote-sensing-images-on-levir-cdRSP-ResNet-50#26F1: 90.10
building-change-detection-for-remote-sensing-images-on-levir-cdRSP-Swin-T#27F1: 90.10
object-detection-in-aerial-images-on-dotaRSP-ViTAEv2-S-FPN-ORCN#36mAP: 77.72%
object-detection-in-aerial-images-on-dotaIMP-ViTAEv2-S-FPN-ORCN#39mAP: 77.38%
object-detection-in-aerial-images-on-dotaRSP-ResNet-50-FPN-ORCN#46mAP: 76.50%
object-detection-in-aerial-images-on-dotaRSP-Swin-T-FPN-ORCN#48mAP: 76.12%