Multiscale Deep Network with Centerness-Aware Loss for Salient Object Detection

Table 3

Comparison with different λ. When λ = 0.50, the proposed model achieves the best results.

	DUT-OMRON			DUTS			ECSSD			PASCAL-S			HKU-IS
	Fm	MAE	Sm	Fm	MAE	Sm	Fm	MAE	Sm	Fm	MAE	Sm	Fm	MAE	Sm
λ = 0.25	0.761	0.054	0.846	0.825	0.038	0.892	0.918	0.036	0.928	0.837	0.067	0.862	0.907	0.030	0.923
λ = 0.50	0.791	0.049	0.860	0.857	0.034	0.900	0.934	0.032	0.933	0.858	0.061	0.870	0.923	0.025	0.929
λ = 0.75	0.777	0.051	0.847	0.839	0.038	0.887	0.926	0.034	0.926	0.849	0.065	0.861	0.913	0.029	0.920
λ = 1.00	0.778	0.051	0.845	0.844	0.036	0.887	0.927	0.035	0.924	0.845	0.069	0.853	0.917	0.028	0.919