README.md

Weight Standardization

Introduction

@article{weightstandardization,
  author    = {Siyuan Qiao and Huiyu Wang and Chenxi Liu and Wei Shen and Alan Yuille},
  title     = {Weight Standardization},
  journal   = {arXiv preprint arXiv:1903.10520},
  year      = {2019},
}

Results and Models

Faster R-CNN

Backbone	Style	Normalization	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Download
R-50-FPN	pytorch	GN+WS	1x	5.9	11.7	39.7	-	model | log
R-101-FPN	pytorch	GN+WS	1x	8.9	9.0	41.7	-	model | log
X-50-32x4d-FPN	pytorch	GN+WS	1x	7.0	10.3	40.7	-	model | log
X-101-32x4d-FPN	pytorch	GN+WS	1x	10.8	7.6	42.1	-	model | log

Mask R-CNN

Backbone	Style	Normalization	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Download
R-50-FPN	pytorch	GN+WS	2x	7.3	10.5	40.6	36.6	model | log
R-101-FPN	pytorch	GN+WS	2x	10.3	8.6	42.0	37.7	model | log
X-50-32x4d-FPN	pytorch	GN+WS	2x	8.4	9.3	41.1	37.0	model | log
X-101-32x4d-FPN	pytorch	GN+WS	2x	12.2	7.1	42.1	37.9	model | log
R-50-FPN	pytorch	GN+WS	20-23-24e	7.3	-	41.1	37.1	model | log
R-101-FPN	pytorch	GN+WS	20-23-24e	10.3	-	43.1	38.6	model | log
X-50-32x4d-FPN	pytorch	GN+WS	20-23-24e	8.4	-	42.1	38.0	model | log
X-101-32x4d-FPN	pytorch	GN+WS	20-23-24e	12.2	-	42.7	38.5	model | log

Note:

• GN+WS requires about 5% more memory than GN, and it is only 5% slower than GN.
• In the paper, a 20-23-24e lr schedule is used instead of 2x.
• The X-50-GN and X-101-GN pretrained models are also shared by the authors.