Weight Standardization

25 Mar 2019Siyuan QiaoHuiyu WangChenxi LiuWei ShenAlan Yuille

In this paper, we propose Weight Standardization (WS) to accelerate deep network training. WS is targeted at the micro-batch training setting where each GPU typically has only 1-2 images for training... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Object Detection COCO minival Mask R-CNN-FPN (ResNeXt-101, GN+WS) box AP 43.12 # 26
AP50 64.15 # 13
AP75 47.11 # 15
APS 25.49 # 19
APM 47.19 # 14
APL 56.39 # 20
Instance Segmentation COCO minival Mask R-CNN-FPN (ResNeXt-101, GN+WS) mask AP 38.34 # 12
AP50 61.07 # 3
AP75 40.82 # 3
APL 56.08 # 2
APM 41.73 # 2
APS 18.32 # 2