UniConvNet: Expanding Effective Receptive Field while Maintaining Asymptotically Gaussian Distribution for ConvNets of Any Scale

Open paper

Benchmark	Model	Rank	Results
object-detection-on-coco-val2017	UniConvNet-L (IN-22K, Cascade Mask R-CNN 3x)	#62	box AP: 56.6AP50: 75.6AP75: 61.8Params (M): 254.8
object-detection-on-coco-val2017	UniConvNet-L (IN-22K, Cascade Mask R-CNN 1x)	#73	box AP: 55.7AP50: 74.4AP75: 60.4Params (M): 254.8
object-detection-on-coco-val2017	UniConvNet-B (Mask R-CNN 3x)	#134	box AP: 51.2AP50: 72.2AP75: 56.1Params (M): 118.0
object-detection-on-coco-val2017	UniConvNet-T (Mask R-CNN 3x)	#152	box AP: 50.1AP50: 71.0AP75: 54.8Params (M): 50.0