【Portrait分割】BANet:Boundary-Aware Network for Fast and High-Accuracy Portrait Segmentation
文章目录
- Abstract
- 1、Background
- 2、Method of BANet
- 2.1、Network Architecture
- Boundary Feature Mining Branch
- 2.2、Loss Function
Abstract
与通用语义分割任务相比,Portrait分割需要更高的精度
和更快的速度
。
Boundary-Aware Network (BANet) 选择性提取边界区域的细节信息获得高质量的分割结果,且可做到实时分割( ≥ 25FPS)
同时BANet设计了一个refine loss对网络中的图像级梯度信息进行监督。
BANet is an efficient network with only 0.62MB
parameters, it achieves 43 fps on 512 × 512 images
with high-quality results which are finer than annotations
.
1、Background
精细的边界细节信息丢失问题主要有以下两个原因造成:
一方面,深度学习模型的性能非常依赖于训练数据。然后target通常用polygons来标注或通过KNN-matting生成。故像头发丝这种极其精细的边界细节很难被标注出来。
另一方面,传统语义分割任务主要解决复杂场景中的intra-class consistency and the inter-class distinction
问题。Portrait分割属于二分类问题,传统语义分割模型并不合适。
2、Method of BANet
In the task of portrait segmentation, no-boundary area
needs a large receptive field
to make prediction with global context information
, while boundary area
needs small receptive field to focus on local feature contrast
. Hence these two areas need to be treated independently.
In this paper, we propose a boundary attention mechanism and a weighted lossfunction to deal with boundary area and no-boundary area separately.
2.1、Network Architecture
Semantic Branch: 获得比较大的感受野,提高对非边界区域的分割。channel数最大仅为64;
Fusion Part: 采用BiSeNet中的FFM模块,channel attention mechanism
Boundary Feature Mining Branch
semantic branch的输出先通过1 × 1 conv
映射到一个通道,然后再上采样到原图大小,作为boundary attention map
。
BA loss引导boundary attention map
定位边界区域。
Extraction of semantic boundary forces the network to learn a feature with strong inter-class distinction ability.
BA loss的target并不需要手动标注。先使用Canny边缘检测器检测portrait annotation(ground truth),然后检测结果作为BA loss的target
。实现可参考:Canny
最后,输入图像与attention map拼接得到一个4维的特征图。
2.2、Loss Function
未完待续…
【Portrait分割】BANet:Boundary-Aware Network for Fast and High-Accuracy Portrait Segmentation相关推荐
- 时序动作检测《BSN: Boundary Sensitive Network for Temporal Action Proposal Generation》
时序动作检测SSAD<Single Shot Temporal Action Detection>_程大海的博客-CSDN博客_时序动作检测 时序动作检测<BSN: Boundary ...
- 【论文阅读】Gait Quality Aware Network: Toward the Interpretability of Silhouette-Based Gait Recognition
Gait Quality Aware Network: Toward the Interpretability of Silhouette-Based Gait Recognition 摘要 Intr ...
- 【步态识别】GQAN步态质量感知网络 算法学习《Gait Quality Aware Network: Toward the Interpretability of Silhouette-Based》
目录 1. 论文&代码源 2. 论文亮点 3. 框架结构 3.1 FQBlock(Frame Quality Block) 3.2 PQBlock(Part Quality Block) 3. ...
- 目标检测--A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection
A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection ECCV2016 https://g ...
- 快速多尺度人脸检测--Multi-Scale Fully Convolutional Network for Fast Face Detection
Multi-Scale Fully Convolutional Network for Fast Face Detection BMVC 2016 如何能够快速的实现多尺度人脸检测了? 本文的思路是 ...
- PBRNet:Progressive Boundary Refinement Network for Temporal Action Detection (AAAI 2020)
PBRNet:Progressive Boundary Refinement Network for Temporal Action Detection AAAI 2020 中国科学技术大学 欢迎感兴 ...
- Drafting and Revision: Laplacian Pyramid Network for Fast High-Quality Artistic Style Transfer--T Li
[1] Lin T , Ma Z , Li F , et al. Drafting and Revision: Laplacian Pyramid Network for Fast High-Qual ...
- CVPR-Drafting and Revision: Laplacian Pyramid Network for Fast High-Quality Artistic Style Transfer
[CVPR-2021] Drafting and Revision: Laplacian Pyramid Network for Fast High-Quality Artistic Style Tr ...
- Boundary Sensitive Network (BSN) 源码运行
Boundary Sensitive Network (BSN) 源码运行 BSN论文:https://arxiv.org/abs/1806.02964 BSN源码:https://github.co ...
最新文章
- authy不同账户间不同步_「第七期」shopify产品还能同步到微信小程序销售?看这里...
- SSPL的MongoDB再被抛弃,GUN Health也合流PostgreSQL
- 第0周学习资源阅读感悟
- 29-分数求模(逆元)
- 教你从0到1搭建秒杀系统-防超卖
- 数组按逆向求最大差值的算法
- How is SAP Gateway metadata request converted to XML format transformation
- 360 php offer,审批终于通过了,从面试到拿到奇虎360的offer已经失…
- eclipse的SVN插件设置忽略文件
- Java:File.separator作用相当于 ‘ \ ‘
- 基于Linux内核红黑树的TR069参数解析工具:树形结构+CPE RPC支持
- AD域首次登陆修改密码设置
- 【廖雪峰官方网站/Java教程】多线程(3)
- day-60Django
- SWFUpload 2.5.0版 官方说明文档 中文翻译版
- java实现简单的文字pk的小游戏
- DH算法的简单的Java实现
- Netty权威指南(三)Netty入门应用
- 魔兽世界国服服务器稳定,《魔兽世界》国服大服务器功能实装启动
- 008九九乘法表(详解)
热门文章
- [08-16]绿色精品软件更新 华夏黑客联盟整理
- 【CSS】display常用属性
- Dnt 缓存架构学习后的总结
- 【转】DICOM协议新手入门资料-DICOM协议详细解释!!
- 当腾讯会议提示“未检测到可用摄像头,请插入设备后重试”并且在设备管理器中没有发现摄像头驱动
- 输入法竞逐AI“新赛道”,旧有认知被颠覆后行业走向何方?
- 面试官:脏读,不可重复读,幻读是如何发生的?
- 计算机英语第四版英汉对照,计算机英语常用词英汉对照
- Android大神微博集
- Win10下如何按计划自动运行脚本