Table 5.

Performance comparison between the improved Mask R-CNN backbone and the original one on the Grape-A test set.

Backbone	ResNet50-FPN		ResNet50-FPN-ED
	Box	Seg	Box	Seg
$AP$	59.90	61.00	62.50	63.20
$A{P}_{IoU=0.5}$	81.90	84.70	82.80	86.20
$A{P}_{IoU=0.75}$	63.50	71.20	66.60	71.70
$A{R}_{\max =1}$	17.40	17.00	17.80	17.30
$A{R}_{\max =10}$	64.50	64.20	67.10	66.90
$A{R}_{\max =100}$	68.10	68.30	71.00	70.60

The best one is highlighted in bold.