Focal Loss for Dense Object Detection

Benchmark Model Rank Results
2d-object-detection-on-sardet-100kRetinaNet#11box mAP: 47.4
dense-object-detection-on-sku-110kRetinaNet#4AP: 45.5AP75: .389
long-tail-learning-on-coco-mltFocal Loss(ResNet-50)#6Average mAP: 49.46
long-tail-learning-on-voc-mltFocal Loss(ResNet-50)#8Average mAP: 73.88
object-counting-on-carpkRetinaNet (2018)#9MAE: 24.58
object-detection-on-cocoRetinaNet (ResNeXt-101-FPN)#184box mAP: 40.8AP50: 61.1AP75: 44.1APS: 24.1APM: 44.2APL: 51.2
object-detection-on-cocoRetinaNet (ResNet-101-FPN)#204box mAP: 39.1AP50: 59.1AP75: 42.3APS: 21.8APM: 42.7APL: 50.2
object-detection-on-coco-oRetinaNet (ResNet-50)#39Average mAP: 16.6Effective Robustness: 0.18
pedestrian-detection-on-tju-ped-campusRetinaNet#3R (miss rate): 34.73RS (miss rate): 82.99HO (miss rate): 71.31
pedestrian-detection-on-tju-ped-trafficRetinaNet#3R (miss rate): 23.89RS (miss rate): 37.92HO (miss rate): 61.60