Self-Supervised Monocular Depth Estimation Model Based on Global Information Correlation Under Influence of Local Attention

Lei Xiao; Peng Hu; Junjie Ma

doi:10.3788/LOP241870

Journals >Laser & Optoelectronics Progress >Volume 62 >Issue 8 >Page 0815010 > Article

Laser & Optoelectronics Progress
Vol. 62, Issue 8, 0815010 (2025)

Self-Supervised Monocular Depth Estimation Model Based on Global Information Correlation Under Influence of Local Attention

Lei Xiao, Peng Hu^*, and Junjie Ma

Author Affiliations

College of Artificial Intelligence, Anhui University of Science & Technology, Huainan 232001, Anhui , China

show less

DOI: 10.3788/LOP241870 Cite this Article Set citation alerts

Lei Xiao, Peng Hu, Junjie Ma. Self-Supervised Monocular Depth Estimation Model Based on Global Information Correlation Under Influence of Local Attention[J]. Laser & Optoelectronics Progress, 2025, 62(8): 0815010 Copy Citation Text

show less

Fig. 1. Overall framework of proposed network

Download full size

Fig. 2. Internal structure of the DepthNet

Download full size

Fig. 3. Schematic diagram of the internal structure for the joint DSConv & Transformer block

Download full size

Fig. 4. Qualitative comparison between the proposed method and other methods on the KITTI dataset

Download full size

Fig. 5. Qualitative comparison of the proposed method and other methods on the Cityscapes dataset

Download full size

Fig. 6. Model complexity and speed evaluation. (a) Params; (b) FLOPs

Download full size

Method	Data	Abs Rel	Sq Rel	RMSE	RMSE log	$δ_{1}$	$δ_{2}$	$δ_{3}$
Monodepth2^［22］	M	0.115	0.903	4.863	0.193	0.877	0.959	0.981
SGDepth^［32］	M+Se	0.113	0.835	4.693	0.191	0.879	0.961	0.981
SAFENet^［33］	M+Se	0.112	0.788	4.582	0.187	0.878	0.963	0.983
R-MSFM6^［24］	M	0.112	0.806	4.704	0.191	0.878	0.960	0.981
PackNet^［34］	M	0.108	0.727	4.426	0.184	0.885	0.963	0.983
HR-Depth^［9］	M	0.109	0.792	4.632	0.185	0.884	0.962	0.983
Ref. ［35］	M	0.106	0.861	4.699	0.185	0.889	0.962	0.982
CADepth^［23］	M	0.105	0.769	4.535	0.181	0.892	0.964	0.983
DIFFNet^［36］	M	0.102	0.749	4.445	0.179	0.897	0.965	0.983
Proposed	M	0.102	0.734	4.435	0.177	0.895	0.966	0.984

Table 1. Quantitative comparison between the proposed method and other monocular depth estimation methods on the KITTI dataset

Method	Data	Abs Rel	Sq Rel	RMSE	RMSE log	$δ_{1}$	$δ_{2}$	$δ_{3}$
Monodepth2^［22］	M	0.155	1.900	6.796	0.209	0.813	0.943	0.979
DIFFNet^［36］	M	0.140	1.571	6.298	0.192	0.837	0.950	0.983
Proposed	M	0.125	1.300	5.790	0.177	0.861	0.958	0.986

Table 2. Quantitative comparison between the proposed method and other monocular depth estimation methods on the Cityscapes dataset

DSConv	SConv	Shuffle	Params /10⁶	Inference time /ms	Abs Rel	Sq Rel	RMSE	RMSE log	$δ_{1}$	$δ_{2}$	$δ_{3}$
			3.36	4.7	0.124	0.901	4.798	0.205	0.849	0.941	0.962
√			4.20	6.0	0.105	0.749	4.501	0.180	0.891	0.960	0.981
		√	3.36	4.8	0.120	0.850	4.602	0.195	0.860	0.953	0.972
	√		7.20	8.5	0.100	0.720	4.400	0.175	0.900	0.970	0.986
√		√	4.20	5.9	0.102	0.734	4.435	0.177	0.895	0.966	0.984

Table 3. Ablation experimental results

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享