Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection

Benchmark Model Rank Results
object-detection-on-cocoGFLV2 (Res2Net-101, DCN, multiscale)#65box mAP: 53.3AP50: 70.9AP75: 59.2APS: 35.7APM: 56.1APL: 65.6
object-detection-on-cocoGFLV2 (Res2Net-101, DCN)#88box mAP: 50.6AP50: 69AP75: 55.3APS: 31.3APM: 54.3APL: 63.5
object-detection-on-cocoGFLV2 (ResNeXt-101, 32x4d, DCN)#98box mAP: 49AP50: 67.6AP75: 53.5APS: 29.7APM: 52.4APL: 61.4
object-detection-on-cocoGFLV2 (ResNet-101-DCN)#106box mAP: 48.3AP50: 66.5AP75: 52.8APS: 28.8APM: 51.9APL: 60.7
object-detection-on-cocoGFLV2 (ResNet-101)#127box mAP: 46.2AP50: 64.3AP75: 50.5APS: 27.8APM: 49.9APL: 57
object-detection-on-cocoGFLV2 (ResNet-50)#144box mAP: 44.3AP50: 62.3AP75: 48.5APS: 26.8APM: 47.7APL: 54.1
object-detection-on-coco-oGFLv2 (R2-101-DCN)#26Average mAP: 25.1Effective Robustness: 2.6