一种结构感知损失的域适应深度估计方法

doi:10.16180/j.cnki.issn1007-7820.2020.12.003

电子科技 ›› 2020, Vol. 33 ›› Issue (12): 12-16.doi: 10.16180/j.cnki.issn1007-7820.2020.12.003

一种结构感知损失的域适应深度估计方法

詹雁,张娟

上海工程技术大学电子电气工程学院,上海 200162

收稿日期:2019-10-08 出版日期:2020-12-15 发布日期:2020-12-22
作者简介:詹雁(1993-),男,硕士研究生。研究方向:计算机视觉。|张娟(1975-),女,博士,副教授。研究方向:计算机图形学、软件测试。
基金资助:
国家自然科学基金(61772328)

A Domain Adaptive Depth Estimation Method for Structural Perception Loss

ZHAN Yan,ZHANG Juan

College of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China

Received:2019-10-08 Online:2020-12-15 Published:2020-12-22
Supported by:
National Natural Science Foundation of China(61772328)

摘要/Abstract

摘要：

针对域适应深度估计方法中域图像间结构差异较大问题,文中提出一种结构感知损失的域适应深度估计方法。该方法通过预训练的卷积神经网络对图像进行特征提取,在特征上进行结构相似性度量,减小了域图像之间的差异性,提高了转换模块的稳定性。该方法使用合成图像深度对和真实图像训练,不需要真实图像的深度标签和物理几何信息。在KITTI数据集上进行实验,深度准确率达到了96.6%,证明该方法可有效提高深度准确率。

关键词: 深度估计, 图像处理, 单目图像, 感知损失, 域适应, 结构相似性

Abstract:

In the domain adaptive depth estimation method, the structural differences between domain images are large. In view of this problem, a domain adaptive depth estimation method for structural perceptual loss is proposed. This method uses pre-trained convolutional neural networks to extract features from images, and measures structural similarity on features, which reduces the difference between domain images and improved the stability of the transform module. This method uses synthetic image depth pairs and real image training, and eliminates the requirement of depth labels and physical geometric information for real images. Experiments on the KITTI dataset achieve a depth accuracy rate of 96.6%, which proves that the method can effectively improve the depth accuracy.

Key words: depth estimation, image processing, monocular image, perceptual loss, domain adaptation, structural similarity index

中图分类号:

TP751

詹雁,张娟. 一种结构感知损失的域适应深度估计方法[J]. 电子科技, 2020, 33(12): 12-16.

ZHAN Yan,ZHANG Juan. A Domain Adaptive Depth Estimation Method for Structural Perception Loss[J]. Electronic Science and Technology, 2020, 33(12): 12-16.

图/表 4

参考文献 16

[1]	张博. 基于Kinect传感器的移动机器人环境检测方法[J]. 电子科技, 2018,31(7):93-95,99.
	Zhang Bo. Kinect sensor based mobile robot environment detection method[J]. Electronic Science and Technology, 2018,31(7):93-95,99.
[2]	郭姝言, 刘桥, 卢进. 一种改进的医学图像深度信息恢复算法[J]. 电子科技, 2016,29(1):48-50,55.
	Guo Shuyan, Liu Qiao, Lu Jin. An improved medical image depth information recovery algorithm[J]. Electronic Science and Technology, 2016,29(1):48-50,55.
[3]	Eigen D, Fergus R. Predicting depth, surface normals and semantic labels with a common multi-scale convolutional architecture[C]. Santiago:International Conference on Computer Vision, 2015.
[4]	Godard C, Mac Aodha O, Brostow G J. Unsupervised monocular depth estimation with left-right consistency[C]. Hawaii:Computer Vision and Pattern Recognition, 2017.
[5]	Zheng C, Cham T J, Cai J. T2net:Synthetic-to-realistic translation for solving single-image depth estimation tasks[C]. Munich:European Conference on Computer Vision(ECCV), 2018.
[6]	Garg R, Bg V K, Carneiro G, et al. Unsupervised cnn for single view depth estimation: Geometry to the rescue[C]. Amsterdam:European Conference on Computer Vision, 2016.
[7]	Xie J, Girshick R, Farhadi A. Deep3D: Fully automatic 2D-to-3D video conversion with deep convolutional neural networks[C]. Amsterdam:European Conference on Computer Vision, 2016.
[8]	Shrivastava A, Pfister T, Tuzel O, et al. Learning from simulated and unsupervised images through adversarial training[C]. Hawaii:Computer Vision and Pattern Recognition, 2017.
[9]	He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]. Las Vegas:Computer Vision and Pattern Recognition, 2016.
[10]	Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[C]. San Diego: International Conference on Learning Representations, 2015.
[11]	Alain H, Ziou D. Image quality metrics: PSNR vs. SSIM[C]. Istanbul:International Conference on Pattern Recognition, 2010.
[12]	Ronneberger O, Fischer P, Brox T. U-net:Convolutional networks for biomedical image segmentation[C]. Munich:International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015.
[13]	Geiger A, Lenz P, Urtasun R. Are we ready for autonomous driving? The kitti vision benchmark suite[C]. Rhode Island:Computer Vision and Pattern Recognition, 2012.
[14]	Gaidon A, Wang Q, Cabon Y, et al. Virtual worlds as proxy for multi-object tracking analysis[C]. Las Vegas:Computer Vision and Pattern Recognition, 2016.
[15]	Kingma D, Ba J. Adam:a method for stochastic optimization[C]. San Diego:International Conference on Learning Representations, 2015.
[16]	Cordts M, Omran M, Ramos S, et al. The cityscapes dataset for semantic urban scene understanding[C]. Las Vegas: Computer Vision an

方法	数据集	Abs	SQ	RMSE	RMSE(log)	δ<1.25³	δ<1.25²	δ<1.25³
文献[3]	K(I,D)	0.190	1.515	7.156	0.270	0.692	0.899	0.958
文献[4]	K(L,R)	0.169	1.080	5.104	0.273	0.740	0.904	0.962
文献[5]	CS,K(L,R)	0.117	0.762	3.972	0.206	0.860	0.948	0.976
文献[6]	V(I,D), K(I)	0.169	1.230	4.717	0.245	0.769	0.912	0.965
基准	V(I,D)	0.521	11.024	10.598	0.473	0.638	0.755	0.835
本文	V(I,D), K(I)	0.167	1.240	4.715	0.230	0.772	0.914	0.966

一种结构感知损失的域适应深度估计方法

A Domain Adaptive Depth Estimation Method for Structural Perception Loss

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 4

参考文献 16

相关文章 15

Metrics

本文评价

推荐阅读 10

[1]	宗圣康,程建鹏,张西良. 基于光斑位置的起重机轨道高度差自动检测方法[J]. 电子科技, 2022, 35(1): 21-28.
[2]	宋章明,贺慧勇,黄跃俊. 扁平漆包线表面缺陷在线检测研究[J]. 电子科技, 2021, 34(5): 72-78.
[3]	闫超,孙占全,田恩刚,赵杨洋,范小燕. 基于深度学习的医学图像分割技术研究进展[J]. 电子科技, 2021, 34(2): 7-11.
[4]	陈文佑,章伟,史晓帆,宋芳. 一种改进ORB特征匹配的半稠密三维重建ORB-SLAM算法[J]. 电子科技, 2021, 34(12): 62-67.
[5]	秦兴,高晓琪,陈滨. 基于压缩卷积神经网络的图像超分辨率算法[J]. 电子科技, 2020, 33(5): 1-8.
[6]	史伟忠,曹卫卫,范彦铭,董佳筠,陈纾,肖昊. 基于FPGA的深空图像实时边缘检测算法与实现[J]. 电子科技, 2020, 33(5): 45-49.
[7]	章天昊,李郝林. 一种基于高斯拟合法的激光光束中心提取方法[J]. 电子科技, 2020, 33(5): 72-76.
[8]	谢智烜,姚红兵,范宁,陈枫. 面向多目标透镜的连通域标记检测算法[J]. 电子科技, 2020, 33(4): 50-54.
[9]	杨云辉. 基于单目视觉的工件定位技术研究[J]. 电子科技, 2019, 32(12): 72-75.
[10]	康秀娟, 施展. 基于图形处理的人体内异物定位测量系统设计[J]. , 2018, 31(1): 71-.
[11]	章佳佳，杨振超，周律. 基于Matlab图像处理的果核分拣系统[J]. , 2017, 30(9): 50-.
[12]	李志森，陈晓荣. 基于Halcon的银行卡卡号识别[J]. , 2017, 30(9): 56-.
[13]	杨振超，章佳佳，周律，吴飞. 基于LabVIEW的马铃薯分拣技术[J]. , 2017, 30(9): 72-.
[14]	那彦，廖萌萌. 基于傅里叶梅林变换的SURF算法[J]. , 2016, 29(7): 55-.
[15]	段炼,郭庚常. 风云卫星云图自动识别追踪MCS的方法[J]. , 2016, 29(4): 116-.