Object Detection on Foggy Cityscapes
58.1mAPDA-Mamba
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DA-MambaArch.=CNN-SSM Hybrid Attention, Inference Mem. Usg.=1307M, FPS=14.1, Training Mem. Usg.=15500M, Training s/iter=0.782s, FLOPs=148G2026.03 | 58.1 | — | — | — | — | — | — | — | — | — | 5.8 | — | |
| Baseline+ViT BlockArch.=Transformer-based Attention, Inference Mem. Usg.=2090M, FPS=10.6, Training Mem. Usg.=19322M, Training s/iter=0.862s, FLOPs=208G2026.03 | 57.5 | — | — | — | — | — | — | — | — | — | 5.2 | — | |
| Diff. Guided DetectorEvaluation Protocol=Domain Adaptation, Backbone=R1012025.06 | 56.6 | 56.9 | 66.2 | 71.7 | 49.1 | 55.6 | 63.1 | 48.9 | 41.1 | — | — | — | |
| DA-ProArch.=Visual-Language Model, Inference Mem. Usg.=2549M, FPS=2.5, Training Mem. Usg.=4034M, Training s/iter=1.472S, FLOPs=243G2026.03 | 55.9 | — | — | — | — | — | — | — | — | — | 3.3 | — | |
| Baseline+Mamba BlockArch.=SSM-based Attention, Inference Mem. Usg.=1792M, FPS=13.0, Training Mem. Usg.=18027M, Training s/iter=0.799s, FLOPs=182G2026.03 | 55 | — | — | — | — | — | — | — | — | — | 2.7 | — | |
| Diff. Guided DetectorEvaluation Protocol=Domain Generalization, Backbone=R1012025.06 | 54.1 | 55.3 | 62.7 | 68.2 | 45.5 | 52.9 | 62 | 48.4 | 37.4 | — | — | — | |
| Baseline+GC BlockArch.=CNN-based Attention, Inference Mem. Usg.=1341M, FPS=14.1, Training Mem. Usg.=15238M, Training s/iter=0.776s, FLOPs=146G2026.03 | 54.1 | — | — | — | — | — | — | — | — | — | 1.8 | — | |
| DATRArch.=Transformer-based Attention, Inference Mem. Usg.=3206M, FPS=5.8, Training Mem. Usg.=21606M, Training s/iter=0.906s, FLOPs=279G2026.03 | 53.4 | — | — | — | — | — | — | — | — | — | 5.3 | — | |
| GDDEvaluation Protocol=Domain Generalization, Backbone=R1012025.06 | 52.5 | 53.8 | 54.2 | 67.5 | 45.6 | 52.1 | 60.8 | 53.9 | 32.4 | — | — | — | |
| CATEvaluation Protocol=Domain Adaptation2025.06 | 52.5 | 66 | 53 | 63.7 | 44.9 | 44.6 | 57.1 | 49.7 | 40.8 | — | — | — | |
| BaselineArch.=CNN-based Backbone, Inference Mem. Usg.=1190M, FPS=15.2, Training Mem. Usg.=14581M, Training s/iter=0.703s, FLOPs=140G2026.03 | 52.3 | — | — | — | — | — | — | — | — | — | — | — | |
| Diff. DetectorEvaluation Protocol=Domain Generalization, Backbone=SD-1.52025.06 | 50.7 | 53 | 55.4 | 68.1 | 42.1 | 51 | 59.9 | 39.4 | 36.4 | — | — | — | |
| DSCAEvaluation Protocol=Domain Adaptation2025.06 | 50.7 | 54.7 | 54.9 | 68.4 | 40.1 | 55.4 | 61 | 35.9 | 35.1 | — | — | — | |
| HTEvaluation Protocol=Domain Adaptation2025.06 | 50.4 | 55.9 | 50.3 | 67.5 | 40.1 | 52.1 | 55.8 | 49.1 | 32.7 | — | — | — | |
| CMTEvaluation Protocol=Domain Adaptation2025.06 | 50.3 | 66 | 51.2 | 63.7 | 41.4 | 45.9 | 55.7 | 38.8 | 39.6 | — | — | — | |
| GDDEvaluation Protocol=Domain Generalization, Backbone=SD-1.52025.06 | 50.1 | 56.2 | 50.4 | 66.7 | 39.9 | 50.2 | 59.5 | 39.9 | 38 | — | — | — | |
| DDTEvaluation Protocol=Domain Adaptation, Backbone=R1012025.06 | 50 | 53.5 | 52.2 | 64.2 | 43.5 | 50.9 | 60 | 42.4 | 33.6 | — | — | — | |
| MTMEvaluation Protocol=Domain Adaptation2025.06 | 48.9 | 54.4 | 47.7 | 67.2 | 38.4 | 51 | 53.4 | 41.6 | 37.2 | — | — | — | |
| UMGAEvaluation Protocol=Domain Adaptation2025.06 | 47.9 | 58 | 43.7 | 64.9 | 38.3 | 47.9 | 50.1 | 45.6 | 34.8 | — | — | — | |
| MICEvaluation Protocol=Domain Adaptation2025.06 | 47.6 | 52.4 | 47.5 | 67 | 40.6 | 50.9 | 55.3 | 33.7 | 33.9 | — | — | — | |
| CIGAREvaluation Protocol=Domain Adaptation2025.06 | 44.9 | 56.6 | 41.3 | 62.1 | 33.7 | 46.1 | 47.3 | 44.3 | 27.8 | — | — | — | |
| SIGMA++Evaluation Protocol=Domain Adaptation2025.06 | 44.5 | 52.2 | 39.9 | 61 | 34.8 | 46.4 | 45.1 | 44.6 | 32.1 | — | — | — | |
| Faster R-CNN (oracle)Architecture=Faster R-CNN, Training=Oracle2022.04 | 41.5 | 48.2 | — | 66.8 | 31.2 | 47.4 | 40.8 | 32.4 | 27.2 | 38.3 | — | — | |
| Fine-tuningMode=Fine-tuning2026.03 | 41.36 | 58.75 | — | — | 31.22 | — | 43.87 | 32.95 | 40.03 | — | — | — | |
| ABRAMode=ABRA2026.03 | 40.54 | 57.24 | — | — | 29.98 | — | 42.27 | 35.09 | 38.1 | — | — | — | |
| Upper BoundBackbone=VGG-162020.03 | 40.3 | 50 | — | 49.7 | 34.7 | 33.2 | 45.9 | 37.4 | 35.6 | 36.2 | — | — | |
| HTCNBackbone=VGG-162020.03 | 39.8 | 47.4 | — | 47.9 | 32.3 | 33.2 | 47.5 | 40.9 | 31.6 | 37.1 | — | — | |
| Oracle2019.10 | 39.2 | 53.3 | — | 58.8 | 35.4 | 37.8 | 48.4 | 15.8 | 25.2 | 39 | — | — | |
| MADEvaluation Protocol=Domain Generalization2025.06 | 38.6 | 44 | 40.1 | 45 | 30.3 | 34.2 | 47.4 | 42.4 | 25.6 | — | — | — | |
| SourceMode=Source2026.03 | 38.25 | 54.77 | — | — | 29.62 | — | 40.25 | 29.4 | 37.23 | — | — | — | |
| SePiCo (DistCL)Architecture=Faster R-CNN2022.04 | 38.1 | 42.8 | — | 57.5 | 29.1 | 43.9 | 41.2 | 26.1 | 25.1 | 39.1 | — | — | |
| SePiCo (BankCL)Architecture=Faster R-CNN2022.04 | 37.8 | 43 | — | 57.9 | 27 | 43.8 | 42.8 | 26.4 | 22.5 | 38.9 | — | — | |
| SePiCo (ProtoCL)Architecture=Faster R-CNN2022.04 | 37.4 | 40.3 | — | 58.2 | 25.5 | 41.9 | 40.4 | 32.6 | 24.9 | 35 | — | — | |
| Progressive Domain Adaptationvariant=progressive2019.10 | 36.9 | 44.1 | — | 54.4 | 29.1 | 36 | 45.5 | 25.8 | 24.3 | 35.9 | — | — | |
| PDAArchitecture=Faster R-CNN2022.04 | 36.9 | 44.1 | — | 54.4 | 29.1 | 36 | 45.5 | 25.8 | 24.3 | 35.9 | — | — | |
| DDTEvaluation Protocol=Domain Generalization, Backbone=SD-1.52025.06 | 36.1 | — | — | — | — | — | — | — | — | — | — | — | |
| Progressive Domain Adaptationvariant=synthetic augment2019.10 | 36.1 | 43.9 | — | 55 | 28.4 | 36.6 | 45.3 | 18.5 | 24.2 | 37.1 | — | — | |
| EveryPixelMattersArchitecture=Faster R-CNN2022.04 | 36 | 41.5 | — | 56.7 | 24.6 | 41.9 | 38.7 | 26.8 | 22.6 | 35.5 | — | — | |
| MTOR*Backbone=ResNet-502020.03 | 35.1 | 38.6 | — | 44 | 28.3 | 30.6 | 41.4 | 40.6 | 21.9 | 35.6 | — | — | |
| Diversify & Match2019.10 | 34.6 | 38.4 | — | 44.3 | 28.4 | 30.8 | 40.5 | 34.5 | 27.2 | 32.2 | — | — | |
| DD-MRLBackbone=VGG-162020.03 | 34.6 | 38.4 | — | 44.3 | 28.4 | 30.8 | 40.5 | 34.5 | 27.2 | 32.2 | — | — | |
| Strong-Weak Align2019.10 | 34.3 | 36.2 | — | 43.5 | 30 | 29.9 | 42.3 | 32.6 | 24.5 | 35.3 | — | — | |
| SWDABackbone=VGG-162020.03 | 34.3 | 36.2 | — | 43.5 | 30 | 29.9 | 42.3 | 32.6 | 24.5 | 35.3 | — | — | |
| SW-FasterArchitecture=Faster R-CNN2022.04 | 34.3 | 29.9 | — | 43.5 | 32.6 | 36.2 | 35.3 | 42.3 | 30 | 24.5 | — | — | |
| MAFBackbone=VGG-162020.03 | 34 | 39.9 | — | 43.9 | 29.2 | 28.2 | 39.5 | 33.3 | 23.8 | 33.9 | — | — | |
| Selective Align2019.10 | 33.8 | 39 | — | 48.5 | 28 | 33.5 | 38 | 23.3 | 26.5 | 33.6 | — | — | |
| SCDABackbone=VGG-162020.03 | 33.8 | 39 | — | 48.5 | 28 | 33.5 | 38 | 23.3 | 26.5 | 33.6 | — | — | |
| DIDNEvaluation Protocol=Domain Generalization2025.06 | 33.4 | 35.7 | 33.1 | 49.3 | 24.8 | 31.8 | 38.4 | 26.5 | 27.7 | — | — | — | |
| FSDREvaluation Protocol=Domain Generalization2025.06 | 31 | 36.6 | 34.1 | 43.3 | 27.1 | 31.2 | 44.4 | 11.9 | 19.3 | — | — | — | |
| Param∆Mode=Param∆2026.03 | 28.29 | 50.23 | — | — | 20.59 | — | 18.53 | 21.7 | 30.42 | — | — | — | |
| Zero shotMode=Zero shot2026.03 | 27.66 | 48.63 | — | — | 23.2 | — | 18.96 | 16.31 | 31.2 | — | — | — | |
| FRCNN in the wild2019.10 | 27.6 | 35.3 | — | 40.5 | 20 | 25 | 31 | 20.2 | 22.1 | 27.1 | — | — | |
| DA-FasterBackbone=VGG-162020.03 | 27.6 | 35.3 | — | 40.5 | 20 | 25 | 31 | 20.2 | 22.1 | 27.1 | — | — | |
| DA-FasterArchitecture=Faster R-CNN2022.04 | 27.6 | 35.3 | — | 40.5 | 20 | 25 | 31 | 20.2 | 22.1 | 27.1 | — | — | |
| Progressive Domain Adaptationvariant=w/o synthetic2019.10 | 26.9 | 26.9 | — | 46.1 | 20.8 | 30.2 | 37.9 | 7 | 14.7 | 31.5 | — | — | |
| FACTEvaluation Protocol=Domain Generalization2025.06 | 25.3 | 27.7 | 31.3 | 35.9 | 23.3 | 26.2 | 41.2 | 3 | 13.6 | — | — | — | |
| Source OnlyBackbone=VGG-162020.03 | 20.3 | 22.3 | — | 34.3 | 15.3 | 24.1 | 33.1 | 3 | 4.1 | 26.5 | — | — | |
| Faster R-CNN2019.10 | 19.6 | 17.9 | — | 36.9 | 13.9 | 23.3 | 29.4 | 2.4 | 7.1 | 25.7 | — | — | |
| Faster R-CNNArchitecture=Faster R-CNN2022.04 | 18.8 | 23.8 | — | 27.1 | 14.4 | 17.8 | 23.6 | 9.1 | 11.9 | 22.8 | — | — | |
| Task AnalogyMode=Task Analogy2026.03 | 18.12 | 41.14 | — | — | 10.24 | — | 9.35 | 10.1 | 19.77 | — | — | — | |
| Foreground Query RevivalSFDA Protocol=PETS, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.582 | 53.622 | |
| Foreground Query RevivalSFDA Protocol=LPU, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.176 | 55.177 | |
| Foreground Query RevivalSFDA Protocol=LPLD, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -0.58 | 55.259 | |
| Foreground Query RevivalSFDA Protocol=DDT, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.627 | 58.617 | |
| Full-Target OracleDetector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 61.333 | |
| Hard Query RecoverySFDA Protocol=PETS, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 1.852 | 54.893 | |
| Hard Query RecoverySFDA Protocol=LPU, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -8.656 | 46.345 | |
| Hard Query RecoverySFDA Protocol=LPLD, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -9.542 | 46.297 | |
| Hard Query RecoverySFDA Protocol=DDT, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -5.522 | 52.468 | |
| Label-Guided Teacher UpdateSFDA Protocol=PETS, Plugin Category=Optimization Control, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.393 | 53.433 | |
| Label-Guided Teacher UpdateSFDA Protocol=LPU, Plugin Category=Optimization Control, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -1.378 | 53.623 | |
| Label-Guided Teacher UpdateSFDA Protocol=LPLD, Plugin Category=Optimization Control, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -1.247 | 54.592 | |
| Label-Guided Teacher UpdateSFDA Protocol=DDT, Plugin Category=Optimization Control, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 1.053 | 59.043 | |
| Multi-View Foreground RevivalSFDA Protocol=PETS, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.168 | 53.209 | |
| Multi-View Foreground RevivalSFDA Protocol=LPU, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.818 | 55.819 | |
| Multi-View Foreground RevivalSFDA Protocol=LPLD, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.297 | 56.136 | |
| Multi-View Foreground RevivalSFDA Protocol=DDT, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.592 | 58.582 | |
| Multi-View Hard RecoverySFDA Protocol=PETS, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -7.495 | 45.546 | |
| Multi-View Hard RecoverySFDA Protocol=LPU, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -10.861 | 44.14 | |
| Multi-View Hard RecoverySFDA Protocol=LPLD, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -10.856 | 44.983 | |
| Multi-View Hard RecoverySFDA Protocol=DDT, Plugin Category=Object Completion, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -7.29 | 50.7 | |
| Precision-Calibrated ThresholdingSFDA Protocol=PETS, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 3.494 | 56.535 | |
| Precision-Calibrated ThresholdingSFDA Protocol=LPU, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.048 | 55.049 | |
| Precision-Calibrated ThresholdingSFDA Protocol=LPLD, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -0.204 | 55.634 | |
| Precision-Calibrated ThresholdingSFDA Protocol=DDT, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.235 | 58.225 | |
| Prior-Mapped ThresholdingSFDA Protocol=PETS, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.451 | 53.491 | |
| Prior-Mapped ThresholdingSFDA Protocol=LPU, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -0.01 | 54.991 | |
| Prior-Mapped ThresholdingSFDA Protocol=LPLD, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | -0.344 | 55.494 | |
| Prior-Mapped ThresholdingSFDA Protocol=DDT, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.735 | 58.725 | |
| Pure SFDASFDA Protocol=PETS, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 42.66 | |
| Pure SFDASFDA Protocol=LPU, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 47.716 | |
| Pure SFDASFDA Protocol=LPLD, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 48.504 | |
| Pure SFDASFDA Protocol=DDT, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 55.526 | |
| Reliability Score RescalingSFDA Protocol=PETS, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.188 | 53.229 | |
| Reliability Score RescalingSFDA Protocol=LPU, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.017 | 55.018 | |
| Reliability Score RescalingSFDA Protocol=LPLD, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.048 | 55.886 | |
| Reliability Score RescalingSFDA Protocol=DDT, Plugin Category=Pseudo-Label Selection, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | 0.769 | 58.759 | |
| RTSMSFDA Protocol=PETS, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 53.04 | |
| RTSMSFDA Protocol=LPU, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 55.001 | |
| RTSMSFDA Protocol=LPLD, Detector=DETA, Label budget=5%2026.06 | — | — | — | — | — | — | — | — | — | — | — | 55.838 |