point_rend

PointRend

Introduction

[ALGORITHM]

@InProceedings{kirillov2019pointrend,
  title={{PointRend}: Image Segmentation as Rendering},
  author={Alexander Kirillov and Yuxin Wu and Kaiming He and Ross Girshick},
  journal={ArXiv:1912.08193},
  year={2019}
}

Results and models

Backbone	Style	Lr schd	Mem (GB)	Inf time (fps)	box AP	mask AP	Config	Download
R-50-FPN	caffe	1x	4.6		38.4	36.3	config	model | log
R-50-FPN	caffe	3x	4.6		41.0	38.0	config	model | log

Note: All models are trained with multi-scale, the input image shorter side is randomly scaled to one of (640, 672, 704, 736, 768, 800).