Self-Supervised Stereo Matching Algorithm Based on Common View

Fig. 3. Comparison of evaluation results on KITTI2012 training set with randomly initializing model during training process. (a) Training loss; (b) D1; (c) EPE; (d) RPE

Download full size

Fig. 4. Comparison of evaluation results on KITTI2012 training set with pre-training model during fine tuning process. (a) Training loss; (b) D1; (c) EPE; (d) RPE

Download full size

Fig. 5. Two sets of images selected from KITTI2015 training set. (a) Left view image in first set. (b) predicted disparity in first set. (c) left view image in second set. (d) predicted disparity in second set

Download full size

Fig. 6. Comparison of practical application effects with pre-training model after fine tuning. (a) Error map in first set; (b) predicted disparity in first set; (c) error map in second set; (d) predicted disparity in second set

Download full size

Table 1. Running time of different disparity prediction models
View table
Table 1. Running time of different disparity prediction models
Image size /pixel×pixel Running time /s
DispNet^[19] DispNetCorr^[19] GC-Net^[22] iResNet^[21] PSMnet^[24]
375×1242 0.074 0.100 6.848 0.375 3.462
480×960 0.074 0.097 6.527 0.339 3.397

Table 2. Comparison of evaluation results on different datasets after training with randomly initializing model

View table

Table 2. Comparison of evaluation results on different datasets after training with randomly initializing model

Method	KITTI2012 training set			KITTI2015 training set
Method	D1 /%	EPE /pixel	RPE /pixel	D1 /%	EPE /pixel	RPE /pixel
Method in Ref. [19]	14.78	2.37	13.91	4.99	0.91	13.40
Method in Ref. [27]	14.53	2.90	11.75	11.39	1.95	9.92
Method in Ref. [27]-M	10.40	2.71	11.38	6.60	1.46	9.46
Method in Ref. [31]	12.98	2.69	11.13	10.80	1.89	9.48
Method in Ref. [31]-M	10.27	2.73	11.24	7.55	1.60	9.35
Proposed method	10.13	2.72	11.37	6.78	1.64	9.64

Table 3. Comparison of evaluation results on different datasets with pre-training model after fine tuning

View table

Table 3. Comparison of evaluation results on different datasets with pre-training model after fine tuning

Method	KITTI2012 training set			KITTI2015 training set
Method	D1 /%	EPE /pixel	RPE /pixel	D1 /%	EPE /pixel	RPE /pixel
Pre-training	8.07	1.53	12.13	8.06	2.27	11.78
Method in Ref. [19]	5.71	1.31	12.74	2.15	0.69	12.75
Method in Ref. [27]	9.26	2.25	11.29	7.85	1.73	9.71
Method in Ref. [27]-M	7.29	1.70	10.84	5.91	1.31	9.46
Method in Ref. [31]	8.21	1.96	10.74	7.28	1.62	9.27
Method in Ref. [31]-M	7.30	1.59	10.57	5.95	1.33	9.21
Proposed method	6.86	1.61	10.64	5.96	1.35	9.31

Tools

Get Citation

Copy Citation Text

Yufeng Wang, Hongwei Wang, Chen Wu, Yu Liu, Yuwei Yuan, Jicheng Quan. Self-Supervised Stereo Matching Algorithm Based on Common View[J]. Acta Optica Sinica, 2019, 39(2): 0215004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Jul. 20, 2018

Accepted: Oct. 10, 2018

Published Online: May. 10, 2019

The Author Email:

DOI:10.3788/AOS201939.0215004

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Running time of different disparity prediction models

Table 1. Running time of different disparity prediction models

Table 2. Comparison of evaluation results on different datasets after training with randomly initializing model

Table 2. Comparison of evaluation results on different datasets after training with randomly initializing model

Table 3. Comparison of evaluation results on different datasets with pre-training model after fine tuning

Table 3. Comparison of evaluation results on different datasets with pre-training model after fine tuning