Results on object detection datasets

ESPNetv2 models

s	Data	Image Size	FLOPs (in billion)	Params (in million)	mAP @ 0.50:0.95	map @ 0.50	mAP @ 0.75	Link
2.0	COCO	300x300	1.1	5.81	20.80	37.05	20.36	here
2.0	COCO	512x512	3.2	5.81	24.54	42.04	25.21	here

Below table compares the results on the PASCAL VOC 2007 dataset. It is important to NOTE that we do not use COCO data for below models.

s	Data	Image Size	FLOPs (in billion)	Params (in million)	mAP	Link
2.0	07+12	300x300	0.9	4.75	70.27	here
2.0	07+12	512x512	2.5	4.75	75.01	here

See below for class-wise mAP