仔细研究域名转换：语义一致域适应的类别级别对手.pdf

poPoq

190

11页

0次

2021-05-02

50墨值下载

Taking A Closer Look at Domain Shift:

Category-level Adversaries for Semantics Consistent Domain Adaptation

Yawei Luo

1,2

, Liang Zheng

, Tao Guan

, Junqing Yu

, Yi Yang

School of Computer Science & Technology, Huazhong University of Science & Technology

CAI, University of Technology Sydney

Research School of Computer Science, Australian National University

Abstract

We consider the problem of unsupervised domain adapta-

tion in semantic segmentation, in which the source domain is

fully annotated, and the target domain is unlabeled. The key

in this campaign consists in reducing the domain shift, i.e.,

enforcing the data distributions of the two domains to be sim-

ilar. A popular strategy is to align the marginal distribution

in the feature space through adversarial learning. However,

this global alignment strategy does not consider the local

category-level feature distribution. A possible consequence

of the global movement is that some categories which are

originally well aligned between the source and target may

be incorrectly mapped. To address this problem, this paper

introduces a category-level adversarial network, aiming to

enforce local semantic consistency during the trend of global

alignment. Our idea is to take a close look at the category-

level data distribution and align each class with an adaptive

adversarial loss. Speciﬁcally, we reduce the weight of the

adversarial loss for category-level aligned features while

increasing the adversarial force for those poorly aligned. In

this process, we decide how well a feature is category-level

aligned between source and target by a co-training approach.

In two domain adaptation tasks, i.e., GTA5

→

Cityscapes

and SYNTHIA

→

Cityscapes, we validate that the proposed

method matches the state of the art in segmentation accuracy.

1. Introduction

Semantic segmentation aims to assign each pixel of a

photograph to a semantic class label. Currently, the achieve-

ment is at the price of large amount of dense pixel-level

annotations obtained by expensive human labor [

]. An

alternative would be resorting to simulated data, such as com-

puter generated scenes [

], so that unlimited amount

of labels are made available. However, models trained with

the simulated images do not generalize well to realistic do-

Source sample, class A

Target sample, class A

Source sample, class B

Target sample, class B

(a) Classical adversarial loss

(b) Self-adaptive adversarial loss

Classifier boundary

Adversarial loss

Source

Target

Figure 1. (Best viewed in color.) Illustration of traditional and the

proposed adversarial learning. The size of the solid gray arrow

represents the weight of the adversarial loss. (a) Traditional adver-

sarial learning ignores the semantic consistency when pursuing the

marginal distribution alignment. As a result, the global movement

might cause the well-aligned features (class A) to be mapped onto

different distributions. (b) The proposed self-adaptive adversar-

ial learning re-weights the adversarial loss for each feature by a

local alignment score. Our method reduces the inﬂuence of the

adversaries when discovers a high semantic alignment score on a

feature, and vice versa. As is shown, the proposed strategy achieves

category-level distribution alignment for both class A and class B.

mains. The reason lies in the different data distributions

of the two domains, typically known as domain shift. To

address this issue, domain adaptation approaches are pro-

posed to bridge the gap between the source and target do-

mains. In practice, unsupervised domain adaptation (UDA)

is concerned and studied more commonly since it minimizes

human labor. Under the UDA setting, a majority of recent

methods [

] aim to align the feature distributions

arXiv:1809.09478v2 [cs.CV] 27 Sep 2018

加入“知识星球行业与管理资源”库，免费下载报告合集

每月上传分享2000+份最新行业资源（涵盖科技、金融、教育、互联网、

房地产、生物制药、医疗健康等行研报告、科技动态、管理方案等）；

2. 免费下载资源库已存行业报告。

3. 免费下载资源库已存国内外咨询公司管理方案与企业运营制度等。

4. 免费下载资源库已存科技方案、论文、报告及课件。

微信扫码二维码，免费报告轻松领

微信扫码加入“知识星球行业与管理资源”，

获取更多行业报告、管理文案、大师笔记

加入微信群，每日获取免费3+份报告

扫一扫二维码，添加客服微信（微信号：Teamkon2）

添加好友请备注：姓名+单位+业务领域

3. 群主将邀请您进专业行业报告资源群

of 11

50墨值下载

自动驾驶

关注

评论