Authors | Architecture | Training dataset | Input | G. Arch | D. Arch | Loss function | Optimizer | Batch size | Output |
---|---|---|---|---|---|---|---|---|---|
Kitchen [78] | DCGAN | 330 Patches | 25-d Noise vector | Transposed CNN | CNN | CE | Adam | 200 | 16 x 16 x 3 |
Hu [79] | ProstateGAN | 1490 Diffusion images | 100-d Noise vector | Transposed CNN | CNN | Conditional GAN loss | Adam | 64 | 32 x 32 |
Wang [81] | StitchAD-GAN | 483 CS 1942 non-CS | 128-d Noise vector | Transposed CNN | Two CNNs | W-distance & JSD | - | - | 64 x 64 |
Yang [82] | Semi-supervised Sequential GAN | 483 CS | 128-d Noise vector & Encodings of real data | Decoder & image translator | CNN | W-distance & L1 | Adam | 32 | 64 x 64 |
Wang [83] | Semi-supervised sequential GAN with StitchLayer | 483 CS 1942 non-CS | 128-d Noise vector & Encodings of real data | Decoder & U-Net translator | CNN | W-distance & L1 & JSD | Adam | 32 | 64 x 64 |
F-Quilez [84] | DCGAN/ pix2pix | 50 | 100-d Noise vector/ Synthetic mask | Transposed CNN/ U-Net | CNN/ PatchGAN | BCE/ pix2pix loss | Adam | 32/ 1 | 256 x 256 |
Yu et [85] | CapGAN | 24.000 patches each modality | 100-d Noise vector | Transposed CNN | Capsule Network | LSE | Adam | - | 35 x 35 |
Yan [88] | PPWGAN-GP | up-to 1,688 | Noise vector | Transposed CNN | CNN | Custom | Adam | 24 | - |
Gao [89] | DCGAN | G1: 547, G2: 1265 G3: 164 (patches to PNG) | 100-d Noise vector | Transposed CNN | CNN | - | Adam | 64 | 56 x 56 |
Gao [90] | DCGAN | 10293 (patches to PNG) | 100-d Noise vector | Transposed CNN | CNN | - | Adam | 64 | 88 x 88 |
Haarburger [91] | DCGAN/ WGAN | 401.525 patches | Noise vector | Resize-Convolution/ Transposed CNN | CNN | GAN loss/ W-distance | Adam/ RMS-prop | 64 | 64 x 64 |
Sun [92] | MM-GAN | 210 HGG 75 LGG | Label maps | 3D U-Net | 3D CNN | LSE | Adam | 1 | 200 x 160 x 150 |