Table 1. Effect of the number of parallel 3 × 3 convolutional operations (k) in MDConv.

Method	Backbone	k	mAP		mAP50		mAP75
			SIXray10	CargoX	SIXray10	CargoX	SIXray10	CargoX
Cascade R-CNN [33]	ResNeXt-101 w FPN	1	52.1	70.1	78.4	95.9	59.9	81.1
	ResNeXt-101 w FPN	2	54.9	71.9	78.5	96.5	64.1	81.7
	ResNeXt-101 w FPN	3	55.4	72.5	79.6	96.5	63.2	82.2
	ResNeXt-101 w FPN	4	54.9	68.3	79.3	95.5	62.4	77.6