Table 6. Results with different loss function on Cityscapes, using models trained on SIM10K (%).
Method | L multi-img | L multi-ins | L multi-cst | Car AP |
---|---|---|---|---|
Faster R-CNN(baseline) | 34.3 | |||
DA Model [9] | 39.4 | |||
SLNO-S5(ours) | 41.5 | |||
SLNO-S6(ours) | ✓ | 41.8 | ||
SLNO-S7(ours) | ✓ | 42.2 | ||
SLNO-S8(ours) | ✓ | ✓ | 42.6 | |
SLNO-S9(ours) | ✓ | ✓ | ✓ | 42.7 |
SLNO(λ = 0.15)(ours) | ✓ | ✓ | ✓ | 43.0 |