SVRreconstruction worse results than reconstruction_GPU2

[hambogo]

I have an old version of reconstruction_GPU2 and it seems to produce better results than SVR. Using --resolution 1.0 and no other options.
Any advice on how to improve SVR results or is the old source code for reconstruction_GPU2 available to rebuild with cuda8?

[bkainz]

I assume the new PSF forumlation does not map 1:1 to your setting. You can always fork the source code and continue developing from there or you can change the PSF to the more general Gaussian blob approximation. For a quick test you can for example fall back to --useCPU, since the CPU implementation only supports the Gaussian formulation. For the GPU version you need to either deactivate USE_SINC_PSF in reconstruction_cuda2.cuh for the previous SVR implementation or adopt the PSF for the PVR implementation in reconConfig.cuh and pointSpreadFunction.cuh.