Progressively Normalized Self-Attention Network for Video Polyp Segmentation

Existing video polyp segmentation(VPS) models typically employ convolutional neural networks (CNNs) to extract features. However, due to their limited receptive fields, CNNs cannot fully exploit the global temporal and spatial information in successive video frames, resulting in false positive segme...

Full description

Saved in:
Bibliographic Details
Published inMedical Image Computing and Computer Assisted Intervention - MICCAI 2021 Vol. 12901; pp. 142 - 152
Main Authors Ji, Ge-Peng, Chou, Yu-Cheng, Fan, Deng-Ping, Chen, Geng, Fu, Huazhu, Jha, Debesh, Shao, Ling
Format Book Chapter
LanguageEnglish
Published Switzerland Springer International Publishing AG 2021
Springer International Publishing
SeriesLecture Notes in Computer Science
Subjects
Colonoscopy
Normalized self-attention
Polyp segmentation
Online AccessGet full text
ISBN3030871924
9783030871925
ISSN0302-9743
1611-3349
DOI10.1007/978-3-030-87193-2_14

Cover

More Information
Summary:Existing video polyp segmentation(VPS) models typically employ convolutional neural networks (CNNs) to extract features. However, due to their limited receptive fields, CNNs cannot fully exploit the global temporal and spatial information in successive video frames, resulting in false positive segmentation results. In this paper, we propose the novel PNS-Net (Progressively Normalized Self-attention Network), which can efficiently learn representations from polyp videos with real-time speed (∼ $$\sim $$ 140fps) on a single RTX 2080 GPU and no post-processing. Our PNS-Net is based solely on a basic normalized self-attention block, equipping with recurrence and CNNs entirely. Experiments on challenging VPS datasets demonstrate that the proposed PNS-Net achieves state-of-the-art performance. We also conduct extensive experiments to study the effectiveness of the channel split, soft-attention, and progressive learning strategy. We find that our PNS-Net works well under different settings, making it a promising solution to the VPS task.
Bibliography:Original Abstract: Existing video polyp segmentation(VPS) models typically employ convolutional neural networks (CNNs) to extract features. However, due to their limited receptive fields, CNNs cannot fully exploit the global temporal and spatial information in successive video frames, resulting in false positive segmentation results. In this paper, we propose the novel PNS-Net (Progressively Normalized Self-attention Network), which can efficiently learn representations from polyp videos with real-time speed (∼\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim $$\end{document}140fps) on a single RTX 2080 GPU and no post-processing. Our PNS-Net is based solely on a basic normalized self-attention block, equipping with recurrence and CNNs entirely. Experiments on challenging VPS datasets demonstrate that the proposed PNS-Net achieves state-of-the-art performance. We also conduct extensive experiments to study the effectiveness of the channel split, soft-attention, and progressive learning strategy. We find that our PNS-Net works well under different settings, making it a promising solution to the VPS task.
G.-P. Ji and Y.-C. Chou—Contributed equally. Code: http://dpfan.net/pnsnet/.
Electronic supplementary materialThe online version of this chapter (https://doi.org/10.1007/978-3-030-87193-2_14) contains supplementary material, which is available to authorized users.
ISBN:3030871924
9783030871925
ISSN:0302-9743
1611-3349
DOI:10.1007/978-3-030-87193-2_14