Table 2. Comparison of the performance of the MDConv module with previous applications of conditional convolutional modules on the backbones of object detectors on the SIXray10 and CargoX datasets.

Method	Backbone	Addition Module	mAP		mAP50		mAP75
			SIXray10	CargoX	SIXray10	CargoX	SIXray10	CargoX
Cascade R-CNN [33]	ResNeXt-101 w FPN	-	52.5	70.2	78.3	95.9	60.6	81.1
	ResNeXt-101 w FPN	SAC [25]	43.0	63.9	67.9	92.8	47.0	71.3
	ResNeSt (s101) [24]	-	51.4	67.1	75.4	94.1	58.6	74.8
	ResNeXt-101 w FPN	DCN [26]	54.2	72.1	78.8	96.4	62.8	82.4
	ResNeXt-101 w FPN	MDconv	55.4	72.5	79.6	96.5	63.2	82.2