Acta Optica Sinica, Volume. 43, Issue 21, 2115003(2023)
Depth Estimation Method of Light Field Based on Attention Mechanism of Neighborhood Pixel
Figures & Tables(14)
Fig. 6. MSE and BP iteration process. (a) MSE iteration process; (b) BP iteration process
Table 1. Comparison of BP between proposed method and other algorithms
View tableTable 1. Comparison of BP between proposed method and other algorithms
Backgammon Dots Pyramids Strips Boxes Cotton Sideboard Avg OFSY_330 4.828 37.670 0.356 18.640 19.246 3.036 10.355 13.447 CAE 3.924 12.401 1.681 7.872 17.885 3.369 9.845 8.140 EPI-refocus 4.305 3.904 0.424 3.922 12.170 0.559 5.955 4.463 EPINET 3.501 2.490 0.159 2.457 12.341 0.447 4.462 3.694 OBER-cross+ANP 3.413 0.974 0.364 3.065 10.759 1.018 5.671 3.609 LFattNET 3.126 1.432 0.195 2.933 11.044 0.272 2.870 3.125 SPO-MO 3.450 2.781 0.050 4.118 15.494 2.161 7.515 5.081 Fast-LFnet 5.138 21.169 0.620 9.442 18.699 0.714 7.032 7.467 OAVC 3.120 69.100 0.830 2.900 16.100 2.550 12.400 15.286 OACC-Net 3.931 1.510 0.157 2.920 10.700 0.312 3.350 3.269 Ours 2.179 0.772 0.365 1.779 12.087 0.497 3.956 3.091
Table 2. Comparison of MSE between proposed method and other algorithms
View tableTable 2. Comparison of MSE between proposed method and other algorithms
Backgammon Dots Pyramids Strips Boxes Cotton Sideboard Avg OFSY_330 7.549 14.756 0.008 7.269 9.561 2.653 2.478 6.325 CAE 6.074 5.082 0.048 3.556 8.424 1.506 0.876 3.652 EPI-refocus 5.553 3.063 0.041 1.870 7.552 0.573 1.609 2.894 EPINET 3.705 1.475 0.007 0.932 5.968 0.197 0.798 1.869 OBER-cross+ANP 4.799 1.757 0.008 1.435 4.750 0.555 0.941 2.035 LFattNET 3.648 1.425 0.004 0.892 3.996 0.209 0.531 1.529 SPO-MO 4.133 3.763 0.009 1.934 10.374 1.329 0.932 3.211 Fast-LFnet 1.488 3.070 0.018 0.231 4.260 0.339 0.742 1.450 OAVC 3.840 16.600 0.040 1.320 6.990 0.600 1.050 4.349 OACC-Net 3.938 1.418 0.004 0.845 2.892 0.162 0.542 1.400 Ours 2.056 0.795 0.010 0.339 3.771 0.260 0.649 1.126
Table 3. Comparison of operation time
View tableTable 3. Comparison of operation time
Algorithms
Avg runtime
Algorithms
Avg runtime
Ours
0.387 SPO-MO
2115.417
LFattNET
5.862
EPI-refocus
72.742
OAVC
4.220
EPINET
2.041
Fast-LFnet
0.624
Table 4. Results of ablation study
View tableTable 4. Results of ablation study
Model MSE /pixel BP /% Training Testing Training Testing Baseline 1.932 2.279 12.087 14.610 Baselines+3D_Conv 1.475 2.012 9.424 11.919 Baseline+Mix_Att 1.352 1.656 5.406 9.825 Full 0.728 1.184 2.838 4.820
Tools
Get Citation
Copy Citation Text
Xi Lin, Yang Guo, Yongqiang Zhao, Naifu Yao. Depth Estimation Method of Light Field Based on Attention Mechanism of Neighborhood Pixel[J]. Acta Optica Sinica, 2023, 43(21): 2115003
Paper Information
Category: Machine Vision
Received: Apr. 7, 2023
Accepted: Jun. 26, 2023
Published Online: Nov. 8, 2023
The Author Email: Zhao Yongqiang (zhaoyq@nwpu.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20