After successful training in previous, we increased complexity of the problem gradually.
- NN trained for multiple bubbles in the target scene with constant Gaussian shaped PSF but PSF varies over the training set
- NN trained for multiple bubbles in the target scene with varying Gaussian shaped PSFs
At this step, we are ready to move to use PSFs from Field2 simulation program instead of Gaussian shaped PSFs. Those PSFs obtained from Field2 are more realistic.
- NN trained for single bubble in the target scene with constant PSF from Field2
- NN trained for multiple bubbles in the target scene with constant PSFs from Field2
- NN trained for single bubble in the target scene with varying PSF from Field2(closeregionPSf, centerPSf, farregionPSF)
- NN trained for multiple bubbles in the target scene with varying PSFs from Field2(closeregionPSf, centerPSf, farregionPSF)
We observe that U-net based NN is able to handle all the complexity levels. Here,there are some examples from the last case:
Example 1:
Example 2:
Example 3:
