Research Article
An Improved Nondestructive Semantic Segmentation Method for Concrete Dam Surface Crack Images with High Resolution
Table 1
Different backbones and architectures and their FLOPs.
| Layer name | Output size | ResNetv2-50 | ResNetv2-50 s | DeepLabv3-DUC ResNetv2-50 | DeepLabv3-DUC+ ResNetv2-50 s |
| conv1 | 112 × 112 | 7 × 7, 64, stride 2 3 × 3 max pool, stride 2 |
| conv2_x | 56 × 56 | | | | |
| conv3_x | 28 × 28 | | | | |
| conv4_x | 14 × 14 | | | | |
| conv5_x | 7 × 7 | | | | | 1 × 1 | Average pool, 2-d fc, softmax | | |
| conv5_d | 14 × 14 | | | | |
| conv6_d | 14 × 14 | | |
|
| 1 × 1 | | | | |
| conv7_d | 14 × 14 | | | | | 14 × 14 | | | | |
| DUC | 224 × 224 | | | Reshape, 14 × 14 × 512 to 224 × 224 × 2, softmax | |
| FLOPs | | 3.87 × 109 [33] | 1.04×109 | 8.72 × 109 | 3.08 × 109 |
|
|