Object Detection on Sim10k to Cityscapes synthetic to real (val)
62.7APOracle
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle2021.03 | 62.7 | |
| OracleTraining=Trained on target dataset labels2018.12 | 53.1 | |
| CDNGlobal=true, Instance=true2020.03 | 49.3 | |
| Proposed (Strong-Weak DA) with FL (y=3)*Global-level alignment (G)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true, Pixel-level alignment (P)=true, Gamma (y)=3, Image Scaling=Shorter side scaled to 10002018.12 | 47.7 | |
| SWDAGlobal=true, Instance=true2020.03 | 47.7 | |
| CDNGlobal=false, Instance=true2020.03 | 45.8 | |
| MeGA-CDA2021.03 | 44.8 | |
| SCDAGlobal=true, Instance=true2020.03 | 44.1 | |
| Selective DA2021.03 | 43 | |
| ATF2021.03 | 42.8 | |
| Proposed (Strong-Weak DA) with FL (y=3)Global-level alignment (G)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true, Gamma (y)=32018.12 | 42.3 | |
| Proposed (Strong-Weak DA) with Focal LossGlobal-level alignment (G)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true, Pixel-level alignment (P)=true2018.12 | 41.5 | |
| CDNGlobal=true, Instance=false2020.03 | 41.2 | |
| MAF2021.03 | 41.1 | |
| Proposed (Strong-Weak DA) with Focal LossGlobal-level alignment (G)=true, Instance-level alignment (I)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true, Pixel-level alignment (P)=true2018.12 | 40.7 | |
| Proposed (Strong-Weak DA) with Focal LossGlobal-level alignment (G)=true, Instance-level alignment (I)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true2018.12 | 40.1 | |
| Strong-Weak2021.03 | 40.1 | |
| DA-Faster*Global-level alignment (G)=true, Instance-level alignment (I)=true, Image Scaling=Shorter side scaled to 10002018.12 | 38.9 | |
| DAFaster2021.03 | 38.9 | |
| Proposed (Strong-Weak DA) with EFLGlobal-level alignment (G)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true, Loss function=Exponential Focal Loss (EFL)2018.12 | 38.7 | |
| Proposed (Strong-Weak DA) with Focal Loss (Zero-padded)Global-level alignment (G)=true, Context vector regularization (CTX)=Zero-padded at output, Local-level alignment (L)=true2018.12 | 38.3 | |
| DA-FasterGlobal=true, Instance=true2020.03 | 38.3 | |
| Proposed (Strong-Weak DA) with Focal LossGlobal-level alignment (G)=true, Context vector regularization (CTX)=true, Local-level alignment (L)=true2018.12 | 38.2 | |
| Proposed (Strong-Weak DA) with Focal LossGlobal-level alignment (G)=true2018.12 | 36.4 | |
| Weak AlignGlobal-level alignment (G)=true, Instance-level alignment (I)=true2018.12 | 35.8 | |
| Faster RCNNAlignment=None2018.12 | 34.6 | |
| Source onlyGlobal=false, Instance=false2020.03 | 34.3 | |
| Source Only2021.03 | 34.3 | |
| DA-FasterGlobal-level alignment (G)=true, Instance-level alignment (I)=true2018.12 | 34.2 | |
| BDC-FasterGlobal-level alignment (G)=true2018.12 | 31.8 |