While it is natural to train the model using single-shot data (because of the turbulence speed that rapidly changes the speckle), a good sanity test is to see if the prediction gets better by doing inference in ensemble mode, i.e. using more than one speckle at the same turbulence for a single prediction.
This can be implemented as a validation test after training.
While it is natural to train the model using single-shot data (because of the turbulence speed that rapidly changes the speckle), a good sanity test is to see if the prediction gets better by doing inference in ensemble mode, i.e. using more than one speckle at the same turbulence for a single prediction.
This can be implemented as a validation test after training.