Table 5 Summary of “DL-post-preprocessing for phase recovery”
Task | Reference | Input | Output | Network | Training dataset | Loss function |
|---|---|---|---|---|---|---|
Noise reduction | Jeon et al.208 | Noisy hologram | Noise-free hologram | U-Net | Sim.: 384,000 pairs | l2-norm and Edge |
Choi et al.209 | Noisy tomogram | Noise-free tomogram | U-Net | Expt.: 455 and 5,057 (unpaired) | Cycle-GAN loss | |
Zhang et al.210 | Noisy wrapped phase | Noise-free wrapped phase | CNN | Sim.: 500 pairs | --- | |
Noisy sine and cosine | Noise-free sine and cosine | ResNet | Sim.: 40,000 and 30,000 pairs | l2-norm | ||
Montresor et al.213 | Noisy sine and cosine | Noise-free sine and cosine | DnCNN | Sim.: 40 pairs (15,360 patches) | l2-norm | |
Noisy sine and cosine | Noise-free sine and cosine | DnCNN | Sim.: 25 pairs and 128 pairs | l2-norm | ||
Fang et al.216 | Noisy real, imaginary | Noise-free real, imaginary | U-Net | Sim.: 4000 pairs | GAN loss | |
Murdaca et al.217 | Noisy real, imaginary, and amplitude | Noise-free real, imaginary, and amplitude | U-Net | Sim.: 5400 pairs | l2-norm | |
Tang et al.219 | Fixed noise matrix | Noise-free phase | U-Net (untrained) | Expt.: 1 | l2-norm, gradient, and variance | |
Resolution enhancement | Liu et al.220 | LR phase and amplitude | HR phase and amplitude | U-Net | Expt.: >50,000 pairs | GAN loss |
Jiao et al.221 | LR phase from DPM | HR phase from SLIM | U-Net | Expt.: >1200 pairs (>100 cells) | l2-norm | |
Butola et al.223 | LR phase | HR phase | U-Net | Expt.: 2355 pairs and 2279 pairs | GAN loss | |
Meng et al.224 | LR phase from SI-DHM | HR phase from SI-DHM | U-Net | Expt.: 3800 pairs | l2-norm | |
Li et al.226 | LR phase | HR phase | U-Net | Expt.: 1680 pairs | l2-norm | |
Gupta et al.227 | LR phase | HR phase | U-Net | Expt.: 700–2000 (unpaired) | Cycle-GAN loss | |
Lim et al.228 | LR 3D RI tomogram | HR 3D RI tomogram | Residual 3D U-Net | Sim.: 1600 pairs | l2-norm | |
Ryu et al.229 | LR 3D RI tomogram | HR 3D RI tomogram | 3D U-Net | Expt.: 217 and 614 pairs | l2-norm | |
Aberration correction | Nguyen et al.234 | Phase | Binary segmentation | U-Net | Expt.: 1836 pairs | Cross entropy |
Ma et al.235 | Hologram | Binary segmentation | U-Net | Expt.: 1000 pairs | Cross entropy | |
Lin et al.236 | Phase and its gradient | Binary segmentation | U-Net and ResNet | Expt.: 1800 pairs | Dice loss | |
Xiao et al.237 | Phase | Zernike coefficient | CNN | Expt.: 10,000 pairs | l2-norm | |
Zhang et al.238 | Aberrated intensity and phase | Phase | U-Net | Sim.: >10,000 groups | l2-norm or l1-norm | |
Tang et al.239 | Fixed vector | Zernike coefficient | MLP (untrained) | Expt. and Sim.: 1 | l2-norm and sparse constraints | |
Phase unwrapping | Wrapped phase | Unwrapped phase | ResNet | Sim.: 7936 pairs | l2-norm | |
Wang et al.245 | Wrapped phase | Unwrapped phase | U-Net and ResNet | Sim.: 30,000 pairs | l2-norm | |
He et al.246 | Wrapped phase | Unwrapped phase | 3D-ResNet | Expt.: --- | --- | |
Ryu et al.247 | Wrapped phase | Unwrapped phase | ReNet | Expt. and Sim.: --- | Total variation and variance | |
Dardikman et al.248 | Wrapped phase | Unwrapped phase | ResNet | Expt.: 7500 pairs | l2-norm | |
Qin et al.249 | Wrapped phase | Unwrapped phase | U-Net and ResNet | Sim.: 30,000 pairs | L1-norm | |
Perera et al.250 | Wrapped phase | Unwrapped phase | U-Net and LSTM | Sim.: 6000 pairs | Total variation and variance | |
Park et al.251 | Wrapped phase | Unwrapped phase | U-Net | Expt.: 5200 pairs | GAN loss | |
Zhou et al.252 | Wrapped phase and wrap count | Unwrapped phase | U-Net and EfficientNet | Sim.: 6000 pairs | l1-norm and residual | |
Xu et al.253 | Wrapped phase | Unwrapped phase | U-Net | Sim.: 6000 pairs | SSIM | |
Zhou et al.254 | Wrapped phase | Unwrapped phase | U-Net | Sim.: 158 and 1036 pairs | GAN loss | |
Xie et al.255 | Wrapped phase | Unwrapped phase | U-Net | Sim.: 17,000 pairs | l2-norm | |
Zhao et al.256 | Wrapped phase and weighted map | Unwrapped phase | U-Net and ResNet | Sim.: 22,500 pairs | l1-norm | |
Liang et al.257 | Wrapped phase | Wrap count | --- | --- | --- | |
Spoorthi et al.258 | Wrapped phase | Wrap count | SegNet | Sim.: 10,000 pairs | Cross entropy | |
Spoorthi et al.259 | Wrapped phase | Wrap count | SegNet and DenseNet | Sim.: 30,000 pairs | Cross entropy and residue and l1-norm | |
Wrapped phase | Wrap count | U-Net | Sim.: 9500 pairs | Cross entropy | ||
Zhang et al.261 | Wrapped phase | Wrap count | DeepLab-V3+ | Sim.: 25,000 pairs | Cross entropy | |
Zhu et al.262 | Wrapped phase | Wrap count | DeepLab-V3+ | Sim.: 20,000 pairs | Cross entropy | |
Wu et al.263 | Wrapped phase | Wrap count | U-Net and FRRNet | Sim.: 12,000 pairs | Cross entropy | |
Zhao et al.264 | Wrapped phase | Wrap count | ResNet | Sim.: 22,000 pairs | Cross entropy | |
Wrapped phase | Wrap count and denoised wrapped phase | Y-Net | Sim.: 2000 pairs | Cross entropy and l2-norm | ||
Zhang et al.267 | Wrapped phase | Wrap count | U-Net and ASPP and PSA | Sim.: 10,000 pairs | Weighted cross entropy | |
Huang et al.278 | Wrapped phase | Wrap count | HRNet | Sim.: 30,000 pairs | Cross entropy | |
Wang et al.279 | Wrapped phase | Wrap count | U-Net, ASPP and EEB | Sim.: --- | Cross entropy | |
Zhou et al.270 | Wrapped count | Wrap count gradient | CNN | Sim.: 52,391 pairs | Cross entropy | |
Wang et al.271 | Wrapped count and quality map | Wrap count gradient | U-Net | Sim.: 164,726 pairs | Cross entropy and dice loss | |
Sica et al.268 | Wrapped count | Wrap count gradient | U-Net | Sim.: >70,000 pairs | Cross entropy, Jaccard distance, and l1-norm | |
Li et al.269 | Wrapped count | Wrap count gradient | U-Net and ResNet | Sim.: 14,100 pairs | Cross entropy | |
Wrapped count | Discontinuity map | CNN and ASPP | Sim.: 8000 pairs | l2-norm, cross entropy, and dice loss | ||
Zhou et al.274 | Residue image | Branch-cut map | CNN | Sim.: 26,928 pairs | Cross entropy |
