Table 2
Evaluation metrics for all presented models in this work.
| Model | Loss function | Learning rate | RMSE (PPF) | MAE (PPF) | COS | CC | SSIM |
|---|---|---|---|---|---|---|---|
| Model 1 (intensity model) | MASE | 0.01 | 1.042 | 0.790 | 0.997 | 0.784 | 0.791 |
| Model 2 (magnetic model) | MASE | 0.01 | 0.961 | 0.745 | 0.998 | 0.806 | 0.799 |
| Model 3 (hybrid model) | MASE | 0.01 | 0.842 | 0.639 | 0.998 | 0.851 | 0.826 |
| FLCT-I | / | / | 1.712 | 1.339 | 0.993 | 0.229 | 0.655 |
| FLCT-M | / | / | 1.669 | 1.077 | 0.995 | 0.517 | 0.669 |
| Original DeepVel | MSE | 0.0001 | 3.174 | 2.566 | 0.980 | 0.592 | 0.485 |
| Retrained DeepVel | MSE | 0.0001 | 1.505 | 1.201 | 0.994 | 0.533 | 0.665 |
Notes. Models 1,2, and 3 are the intensity, magnetic, and hybrid models built on a shallow U-Net architecture. FLCT-I and FLCT-M are the velocity fields derived from the photospheric intensity and vertical magnetic field strength, respectively, using the FLCT method (e.g. Fisher & Welsch 2008). The last two rows are results from the original DeepVel (Asensio Ramos et al. 2017) and the retrained DeepVel (Sect. 4.2). See the main text for the definition of the loss functions and evaluation metrics.
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.