Network | Description | Total Parameters | Training Set Error | Training Set Accuracy | Validation Set Error | Validation Set Accuracy | Stop Learning Epoch |
(a) UNet | Original U shape network with preprocessing dataset | 7,759,686B | 0.07306 | 0.9789 | 0.1218 | 0.9639 | 80 |
(b) UNet_1bn | Add one Batch Normalization | 7,759,690B | 0.04543 | 0.9835 | 0.0505 | 0.9816 | 81 |
(c) sUNet_1bn | Simplified to one convolution each layer & Add one Batch Normalization | 3,832,326B | 0.06771 | 0.9764 | 0.0981 | 0.9626 | 80 |
(d) sUNet_1bn_ dilation12312 | use dilated convolution instead of convolution in sUNet_1bn | 3,832,326B | 0.06220 | 0.9779 | 0.1039 | 0.9664 | 71 |