@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}
}
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).