Micro-Batch Training with Batch-Channel Normalization and Weight Standardization

Benchmark Model Rank Results
instance-segmentation-on-coco-minivalMask R-CNN-FPN (ResNeXt-101, GN+WS)#76mask AP: 38.34AP50: 61.07AP75: 40.82APL: 56.08APM: 41.73
object-detection-on-coco-minivalMask R-CNN-FPN (ResNeXt-101, GN+WS)#144box AP: 43.12AP50: 64.15AP75: 47.11APS: 25.49APM: 47.19