Table 1.
Results of experiments with various neural network architectures on the validation dataset.
| Architecture | C | S | C&S |
|
|---|---|---|---|---|
| C | S | |||
| FCN | 0.0580 | 0.8877 | – | – |
| DeconvNet | 0.0124 | 0.8991 | 0.0128 | 0.8763 |
| U-Net | 0.0112 | 0.9105 | 0.0119 | 0.8866 |
| UNet++ | 0.0146 | 0.8945 | 0.0133 | 0.8679 |
| FPN | 0.0125 | 0.9063 | 0.0119 | 0.8887 |
| PSPNet | 0.0115 | 0.9154 | 0.0119 | 0.8914 |
| UPPNet (sparse) | 0.0122 | 0.9166 | 0.0126 | 0.8857 |
| UPPNet (full) | 0.0116 | 0.9065 | 0.0110 | 0.9033 |
| SUPPNet (synth) | – | – | 0.0092 | 0.9132 |
Notes. S stands for segmentation, C for pseudo-continuum prediction. We report accuracy metrics for the former and mean absolute error for the latter. The last two columns contain the results achieved by models trained in both tasks simultaneously (C and S). The best results are in bold. The results in the last row anticipate the metrics of the final model. These models are described in detail in Sect. 3.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.