We present a new method for efficient high-quality image segmentation of
objects and scenes. By analogizing classical computer graphics methods for
efficient rendering with over- and undersampling challenges faced in pixel
labeling tasks, we develop a unique perspective of image segmentation as a
rendering problem. From this vantage, we present the PointRend (Point-based
Rendering) neural network module: a module that performs point-based
segmentation predictions at adaptively selected locations based on an iterative
subdivision algorithm. PointRend can be flexibly applied to both instance and
semantic segmentation tasks by building on top of existing state-of-the-art
models. While many concrete implementations of the general idea are possible,
we show that a simple design already achieves excellent results. Qualitatively,
PointRend outputs crisp object boundaries in regions that are over-smoothed by
previous methods. Quantitatively, PointRend yields significant gains on COCO
and Cityscapes, for both instance and semantic segmentation. PointRend's
efficiency enables output resolutions that are otherwise impractical in terms
of memory or computation compared to existing approaches.
Описание
[1912.08193] PointRend: Image Segmentation as Rendering
%0 Generic
%1 kirillov2019pointrend
%A Kirillov, Alexander
%A Wu, Yuxin
%A He, Kaiming
%A Girshick, Ross
%D 2019
%K 2019 graphics segmentation
%T PointRend: Image Segmentation as Rendering
%U http://arxiv.org/abs/1912.08193
%X We present a new method for efficient high-quality image segmentation of
objects and scenes. By analogizing classical computer graphics methods for
efficient rendering with over- and undersampling challenges faced in pixel
labeling tasks, we develop a unique perspective of image segmentation as a
rendering problem. From this vantage, we present the PointRend (Point-based
Rendering) neural network module: a module that performs point-based
segmentation predictions at adaptively selected locations based on an iterative
subdivision algorithm. PointRend can be flexibly applied to both instance and
semantic segmentation tasks by building on top of existing state-of-the-art
models. While many concrete implementations of the general idea are possible,
we show that a simple design already achieves excellent results. Qualitatively,
PointRend outputs crisp object boundaries in regions that are over-smoothed by
previous methods. Quantitatively, PointRend yields significant gains on COCO
and Cityscapes, for both instance and semantic segmentation. PointRend's
efficiency enables output resolutions that are otherwise impractical in terms
of memory or computation compared to existing approaches.
@misc{kirillov2019pointrend,
abstract = {We present a new method for efficient high-quality image segmentation of
objects and scenes. By analogizing classical computer graphics methods for
efficient rendering with over- and undersampling challenges faced in pixel
labeling tasks, we develop a unique perspective of image segmentation as a
rendering problem. From this vantage, we present the PointRend (Point-based
Rendering) neural network module: a module that performs point-based
segmentation predictions at adaptively selected locations based on an iterative
subdivision algorithm. PointRend can be flexibly applied to both instance and
semantic segmentation tasks by building on top of existing state-of-the-art
models. While many concrete implementations of the general idea are possible,
we show that a simple design already achieves excellent results. Qualitatively,
PointRend outputs crisp object boundaries in regions that are over-smoothed by
previous methods. Quantitatively, PointRend yields significant gains on COCO
and Cityscapes, for both instance and semantic segmentation. PointRend's
efficiency enables output resolutions that are otherwise impractical in terms
of memory or computation compared to existing approaches.},
added-at = {2019-12-19T17:01:57.000+0100},
author = {Kirillov, Alexander and Wu, Yuxin and He, Kaiming and Girshick, Ross},
biburl = {https://www.bibsonomy.org/bibtex/2350017d9259a795abc6c369b4b745eec/analyst},
description = {[1912.08193] PointRend: Image Segmentation as Rendering},
interhash = {fa809ca4b1636aaa9e48ac2a69c58cf3},
intrahash = {350017d9259a795abc6c369b4b745eec},
keywords = {2019 graphics segmentation},
note = {cite arxiv:1912.08193Comment: Technical Report},
timestamp = {2019-12-19T17:01:57.000+0100},
title = {PointRend: Image Segmentation as Rendering},
url = {http://arxiv.org/abs/1912.08193},
year = 2019
}