Infrared and Laser Engineering, Volume. 54, Issue 6, 20250076(2025)
Multi-scale phase extraction network for structured light based on discrete wavelet transform and attention mechanism
Figures & Tables(13)
Fig. 3. (a) The illustration of multi-scale augmented prediction module; (b) The illustration of bottom up module
Fig. 4. Predication of different networks under fringe pattern with single object. (a), (e), (i)
Fig. 5. Error maps of different networks under fringe pattern with single object. (a)-(d) The error maps of
Fig. 6. Predication of different networks under fringe pattern with multi object. (a), (e), (i)
Fig. 7. Error maps of different networks under fringe pattern with multi object. (a)-(d) The error maps of
Table 1. Generalization Performance Comparison of different networks on the test dataset
View tableView in ArticleTable 1. Generalization Performance Comparison of different networks on the test dataset
Model $ M(x,y) $ $ D(x,y) $ Final phase MAE RMSE PSNR MAE RMSE PSNR UNet 0.003043 0.007330 41.688 0.004956 0.011557 41.328 Att-UNet 0.002870 0.007465 41.309 0.004140 0.011494 41.215 Swin-UNet 0.004689 0.012691 36.582 0.007079 0.019018 36.682 WA-MSPNet 0.002589 0.007050 41.841 0.003927 0.010864 41.694
Table 2. Comparison of different networks on parameters and FLOPs
View tableView in ArticleTable 2. Comparison of different networks on parameters and FLOPs
Model Parameters FLOPs UNet 34.527 M 225.248 G Att-UNet 34.879 M 229.061 G Swin-UNet 97.647 M 97.222 G WA-MSPNet 14.509 M 152.566 G
Table 3. Comparison of inference time for a single image
View tableView in ArticleTable 3. Comparison of inference time for a single image
Model Inference time/ms UNet 19.74 Att-UNet 20.35 Swin-UNet 27.84 WA-MSPNet 17.59
Table 4. Generalization performance comparison of different networks on the test dataset with additive noise
View tableView in ArticleTable 4. Generalization performance comparison of different networks on the test dataset with additive noise
$ M(x,y) $ $ D(x,y) $ Final phase MAE RMSE PSNR MAE RMSE PSNR $ \mathrm{\sigma }=0.01 $ UNet 0.005370 0.009129 39.614 0.008264 0.014285 39.332 Att-UNet 0.005046 0.008985 39.532 0.007549 0.013908 39.385 Swin-UNet 0.006562 0.013653 35.884 0.009771 0.020480 35.976 WA-MSPNet 0.004754 0.008585 39.961 0.007083 0.013161 39.850 $ \sigma =0.05 $ UNet 0.037836 0.047100 25.645 0.057591 0.072598 25.332 Att-UNet 0.030530 0.038783 27.015 0.042372 0.055347 27.634 Swin-UNet 0.021332 0.029109 29.275 0.031097 0.043380 29.426 WA-MSPNet 0.020261 0.026660 30.399 0.029604 0.040011 30.457 $ \sigma =0.10 $ UNet 0.052840 0.064404 23.376 0.077552 0.095627 23.109 Att-UNet 0.058705 0.070645 21.814 0.081091 0.099148 22.558 Swin-UNet 0.049832 0.063202 23.665 0.071577 0.089597 23.774 WA-MSPNet 0.046064 0.056782 24.085 0.066982 0.083292 24.224
Table 5. Configuration of models in ablation study
View tableView in ArticleTable 5. Configuration of models in ablation study
Module A B C D E WA-MSPNet MSAP Π Π Π WT Π Π Π Π HA Π Π
Table 6. Comparison of models in ablation experiments
View tableView in ArticleTable 6. Comparison of models in ablation experiments
$ M(x,y) $ $ D(x,y) $ Final phase MAE RMSE PSNR MAE RMSE PSNR Base model A 0.003184 0.007411 41.350 0.005001 0.011448 41.187 Sub-model B 0.002779 0.007247 41.552 0.004333 0.011205 41.385 Sub-model C 0.002839 0.007703 41.033 0.004351 0.011724 41.018 Sub-model D 0.002699 0.007246 41.563 0.004007 0.011153 41.442 Sub-model E 0.002748 0.007212 41.611 0.004138 0.011159 41.443 WA-MSPNet 0.002589 0.007050 41.841 0.003927 0.010864 41.694
Tools
Get Citation
Copy Citation Text
Jianhua SHANG, Gang WANG, Yang LIU, Haiqin XU, Jiatong SUN. Multi-scale phase extraction network for structured light based on discrete wavelet transform and attention mechanism[J]. Infrared and Laser Engineering, 2025, 54(6): 20250076
Paper Information
Category: Optical imaging, display and information processing
Received: Jan. 24, 2025
Accepted: --
Published Online: Jul. 1, 2025
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20