Laser & Optoelectronics Progress, Volume. 61, Issue 14, 1415004(2024)
Multiview Stereo Reconstruction with Feature Aggregation Transformer
Figures & Tables(11)
Fig. 2. Comparison of convolutional kernel sampling. (a) The convolutional kernel sampling method of standard convolution; (b) (c) (d) convolutional kernel sampling method for deformable convolutions
Fig. 6. Transformer encoder and attention structure. (a) Transformer encoder layer; (b) attention layer
Fig. 8. Comparison of reconstruction results of some models on the Tanks & Temples
Table 1. Quantitative testing results of different methods on DTU
View tableTable 1. Quantitative testing results of different methods on DTU
Method Dacc Dcomp Doverall Furu 0.613 0.941 0.777 Gipuma 0.283 0.873 0.578 COLMAP 0.400 0.664 0.532 MVSNet 0.396 0.527 0.462 Fast-MVSNet 0.336 0.403 0.370 CasMVSNet 0.325 0.385 0.355 UCS-Net 0.338 0.349 0.344 Uni-MVSNet 0.352 0.278 0.315 CVP-MVSNet 0.296 0.406 0.351 CDS-MVSNet 0.352 0.280 0.316 MVSTR 0.356 0.295 0.326 Proposed method 0.335 0.284 0.310
Table 2. Quantitative testing results of different methods on DTU
View tableTable 2. Quantitative testing results of different methods on DTU
Method Int.Mean Intermediate Adv.Mean Advanced Fam. Fra. Hor. Lig. M60 Pan. Pla. Tra. Aud. Bal. Cou. Mus. Pal. Tem. COLMAP 42.14 50.41 22.25 26.63 56.43 44.83 46.97 48.53 42.04 27.24 16.02 25.23 34.7 41.51 18.05 27.94 MVSNet 43.48 55.99 28.55 25.07 50.79 53.96 50.86 47.9 34.69 Fast-MVSNet 47.39 65.18 39.59 34.98 47.81 49.16 46.2 53.27 42.91 PatchmatchNet 53.15 66.99 52.64 43.24 54.87 52.87 49.54 54.21 50.81 32.31 23.69 37.73 30.04 41.8 28.31 32.29 CasMVSNet 56.84 76.37 58.45 46.26 55.81 56.11 54.06 58.18 49.51 31.12 19.81 38.46 29.1 43.87 27.36 28.11 UCS-Net 54.83 76.09 53.16 43.03 54 55.6 51.49 57.38 47.89 CVP-MVSNet 54.03 76.5 47.74 36.34 55.12 57.28 54.28 57.43 47.54 AA-RMVSNet 61.51 77.77 59.53 51.53 64.02 64.05 59.47 60.85 54.9 33.53 20.96 40.15 32.05 46.01 29.28 32.71 CDS-MVSNet 61.58 78.85 63.17 53.04 61.34 62.63 59.06 62.28 52.3 MVSTER 37.53 26.68 42.14 35.65 49.37 32.16 39.19 Proposed method 62.81 78.32 65.21 53.01 62.07 64.48 63.25 61.78 54.36 38.18 27.36 43.37 39.26 52.19 33.46 33.41
Table 3. Comparison of quantitative results of ablation experiments
View tableTable 3. Comparison of quantitative results of ablation experiments
No. Method Dacc Dcomp Doverall 1 Baseline 0.367 0.325 0.346 2 +DCN 0.356 0.304 0.330 3 +SA+CA 0.342 0.288 0.315 4 FAT-MVSNet 0.335 0.284 0.310
Tools
Get Citation
Copy Citation Text
Min Wang, Mingfu Zhao, Tao Song, Weiwei Li, Yuan Tian, Cheng Li, Yu Zhang. Multiview Stereo Reconstruction with Feature Aggregation Transformer[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1415004
Paper Information
Category: Machine Vision
Received: Nov. 22, 2023
Accepted: Dec. 21, 2023
Published Online: Jul. 8, 2024
The Author Email: Tao Song (tsong@cqut.edu.cn)
CSTR:32186.14.LOP232546
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20