In experiments 7, randomized validation set is used.(f,f_0 in pixel coordinates)
In experiments 8, only far field patches are used and randomized validation set used. (f,f_0 in pixel coordinates)
In experiments 9, only far field patches are used and randomized validation set used. (f,f_0 in physical coordinates)
Q1 = \[ L1Loss(z**f_0 – x**f_0) \]
Q2 = \[ MSELossLoss(z**f_0 – x**f_0) \]
where f_0 is a gaussian kernel with sigma =1.
experiments | sigma of f | \(\lambda\) | learning rate | number of epochs | batch size | Q1 | Q2 | Q1 per bubble | Q2 per bubble |
exp1/9 | 1 | 0.01 | 1e-5 | 150 | 1 | 21.066962 | 0.805354 | 0.829031 | 0.031346 |
exp2/9 | 1 | 0.005 | 1e-5 | 150 | 1 | 21.959468 | 0.812720 | 0.863606 | 0.031575 |
exp3/9 | 1.5 | 0.01 | 1e-5 | 150 | 1 | 21.031181 | 0.779206 | 0.827787 | 0.030392 |
exp4/9 | 1.5 | 0.005 | 1e-5 | 150 | 1 | 20.906547 | 0.718547 | 0.823864 | 0.028081 |
exp5/9 | 2 | 0.01 | 1e-5 | 150 | 1 | 22.376564 | 0.872184 | 0.881055 | 0.034107 |
exp6/9 | 2 | 0.005 | 1e-5 | 150 | 1 | 23.040157 | 0.863941 | 0.907826 | 0.033746 |
exp1/8 | 1 | 0.01 | 1e-5 | 150 | 1 | 12.737782 | 0.556178 | 0.501236 | 0.021680 |
exp2/8 | 1 | 0.005 | 1e-5 | 150 | 1 | 13.120187 | 0.516576 | 0.517213 | 0.020148 |
exp3/8 | 1.5 | 0.01 | 1e-5 | 150 | 1 | 13.126150 | 0.544524 | 0.517840 | 0.021331 |
exp4/8 | 1.5 | 0.005 | 1e-5 | 150 | 1 | 13.454322 | 0.492222 | 0.531489 | 0.019310 |
exp5/8 | 2 | 0.01 | 1e-5 | 150 | 1 | 14.191951 | 0.614085 | 0.559143 | 0.024005 |
exp6/8 | 2 | 0.005 | 1e-5 | 150 | 1 | 13.774952 | 0.536625 | 0.543580 | 0.021043 |
exp1/7 | 1 | 0.01 | 1e-5 | 200 | 1 | 19.645459 | 0.981560 | 0.7747 | 0.0384 |
exp2/7 | 1 | 0.005 | 1e-5 | 200 | 1 | 20.352445 | 0.985063 | 0.802987 | 0.038526 |
exp3/7 | 1.5 | 0.01 | 1e-5 | 200 | 1 | 18.815349 | 0.936104 | 0.742511 | 0.036656 |
exp4/7 | 1.5 | 0.005 | 1e-5 | 200 | 1 | 19.907855 | 0.928157 | 0.785045 | 0.036292 |
exp5/7 | 2 | 0.01 | 1e-5 | 200 | 1 | 19.071611 | 1.000624 | 0.753008 | 0.039267 |
exp6/7 | 2 | 0.005 | 1e-5 | 200 | 1 | 19.766309 | 0.954419 | 0.780344 | 0.037464 |