Table 3
Overview of the mean results of the three evaluation methods with clean, noisy and noisy plus additional white noise input maps and for input maps with different mask filling.
| Jet offset [°] | Intensity deviation [%] | Source area ratio | |
|---|---|---|---|
| Noiseless input Vnoiseless | −0.02 ± 1.43 | −4.35 ± 6.66 | 0.96 ± 0.12 |
| Noisy input Vnoisy | −0.01 ± 1.33 | 0.61 ± 12.40 | 0.99 ± 0.13 |
| Noise & white noise Vnoisy&white noise | 0.05 ± 1.96 | 3.49 ± 12.26 | 1.00 ± 0.13 |
| Sampling Density | |||
| 20% | −0.07 ± 3.14 | −8.99 ± 16.26 | 0.85 ± 0.19 |
| 50% | 0.02 ± 1.96 | 6.69 ± 14.45 | 0.92 ± 0.17 |
| 70% | 0.01 ± 3.12 | 11.94 ± 63.60 | 0.91 ± 0.17 |
Notes. A new training set is used for different noise models. The data sets consist of 50 000 training images, 10 000 validation images, and 10 000 test images. For different sampling densities, we do not train a new model but create dedicated test data sets consisting of 10 000 test amplitude and phase maps. The evaluation is done using the deep learning model trained on noisy input data.
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.