Num | Models | Dice | mIOU | PA | Parameters(M) |
|---|
1 | U-net++ | 0.7878 | 0.7344 | 0.9295 | 9.16 |
2 | SegNet | 0.7233 | 0.7026 | 0.9151 | 29.45 |
3 | DeepLabV3+ | 0.7743 | 0.7175 | 0.9236 | 26.72 |
4 | PSPnet | 0.8030 | 0.7864 | 0.9495 | 32.23 |
5 | PA-U-net++ | 0.8188 | 0.8035 | 0.9479 | 25.24 |
- The Dice (dice coefficient) is a set similarity measurement function, which is usually used to calculate the similarity between two sets. You can see in Eq. 14
- The IOU (intersection over union) is the ratio of the intersection and union of the predicted result of a certain category and the true label. You can see in Eq. 15
- For multi-category semantic segmentation, the average intersection over union ratio mean IOU (mIOU) is generally used as the evaluation indicator, that is, the IOU of each category is summed and then averaged
- The PA (pixel accuracy) is the percentage of correct predicted pixels in the total number of pixels. You can see in Eq. 16
- The CPA (category pixel accuracy) is the percentage of pixels whose real tags also belong to category. You can see in Eq. 17
