Table 2.

Obstacle detection comparisons using different feature fusion blocks on the nuScenes test dataset [37]. The top sub-table is about average precision metrics. In addition, the bottom sub-table is about average recall metrics.

Model	Backbone	Scale	$\mathit{AP}$(100)	${\mathit{AP}}^{0.50}$(100)	${\mathit{AP}}^{0.75}$(100)	${\mathit{AP}}^{\mathit{s}}$(100)	${\mathit{AP}}^{\mathit{m}}$(100)	${\mathit{AP}}^{\mathit{l}}$(100)
CAT-FCOS	ResNet-50	800	59.8	85.7	64.4	36.4	57.9	74.8
MUL-FCOS	ResNet-50	800	61.6	86.4	67.5	39.8	60.9	74.5
ADD-FCOS	ResNet-50	800	64.2	87.6	70.9	42.7	62.8	77.4
SAF-FCOS	ResNet-50	800	72.4	90.0	79.3	55.3	70.1	83.1
Model	Backbone	Scale	$\mathit{AR}$ (1)	$\mathit{AR}$ (10)	$\mathit{AR}$ (100)	${\mathit{AR}}^{\mathbf{s}}$ (100)	${\mathit{AR}}^{\mathbf{m}}$ (100)	${\mathit{AR}}^{\mathbf{l}}$ (100)
CAT-FCOS	ResNet-50	800	13.3	58.4	68.1	48.7	66.6	80.6
MUL-FCOS	ResNet-50	800	13.4	60.1	70.2	52.6	69.3	80.9
ADD-FCOS	ResNet-50	800	13.7	61.9	71.9	55.1	70.9	82.5
SAF-FCOS	ResNet-50	800	14.5	68.4	79.0	69.6	77.4	86.5