DetectoRS: Detecting Objects with Recursive Feature Pyramid and Switchable Atrous Convolution

Benchmark Model Rank Results
instance-segmentation-on-cocoDetectoRS (ResNeXt-101-64x4d, multi-scale)#27mask AP: 48.5AP50: 72.0AP75: 53.3APS: 31.6APM: 50.9APL: 61.5
instance-segmentation-on-cocoDetectoRS (ResNeXt-101-32x4d, multi-scale)#32mask AP: 47.1AP50: 71.1AP75: 51.6APS: 30.3APM: 49.5APL: 59.6
object-detection-on-ai-todDetectoRS (ResNet-50-FPN)#4AP: 14.8AP50: 32.8AP75: 11.4APvt: 0.0APt: 10.8APs: 28.3
object-detection-on-cocoDetectoRS (ResNeXt-101-64x4d, multi-scale)#47box mAP: 55.7AP50: 74.2AP75: 61.1APS: 37.7APM: 58.4APL: 68.1
object-detection-on-cocoDetectoRS (ResNeXt-101-32x4d, multi-scale)#52box mAP: 54.7AP50: 73.5AP75: 60.1APS: 37.4APM: 57.3APL: 66.4
object-detection-on-cocoDetectoRS (ResNeXt-101-32x4d, single-scale)#64box mAP: 53.3AP50: 71.6AP75: 58.5APS: 33.9APM: 56.5APL: 66.9
panoptic-segmentation-on-coco-test-devDetectoRS (ResNeXt-101-64x4d, multi-scale)#14PQ: 50PQst: 37.2PQth: 58.5