GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond

Benchmark Model Rank Results
instance-segmentation-on-cocoGCNet (ResNeXt-101 + DCN + cascade + GC r16)#56mask AP: 41.5%
instance-segmentation-on-coco-minivalGCNet (ResNeXt-101 + DCN + cascade + GC r16)#67mask AP: 40.9
object-detection-on-cocoGCNet (ResNeXt-101 + DCN + cascade + GC r4)#105box mAP: 48.4AP50: 67.6AP75: 52.7
object-detection-on-coco-minivalGCNet (ResNeXt-101 + DCN + cascade + GC r16)#93box AP: 47.9AP50: 66.9AP75: 52.2
object-detection-on-coco-minivalGCnet (ResNet-50-FPN, GRoIE)#180box AP: 40.3AP50: 62.4AP75: 44APS: 24.2APM: 44.4APL: 52.5
object-detection-on-coco-oGCNet (RX-101-32x4d-DCN)#25Average mAP: 26.0Effective Robustness: 4.38