You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
So here is the thing. When we do metacal on our simulations as an input to the network, we can either apply metacal on existing images (i.e. which already contain a psf and noise) like in normal metacal, or we can compute the response to a "true shear" introduced in the image before the noise and the PSF.
If we compute the response on existing images, we have to deal with the induced noise correlations that our shearing is doing. And for large shears it will be pretty significant. But, it means that we can use true images from the survey to train the network.
If we compute the response to a true shear, then we don't need to deal with the noise correlations and such. But we can only do this on simulations, because the need to shear the image before applying noise and psf, so not on the actual survey images.
The clean thing we might want to do is this second option, at the cost of not being able to directly use survey images :-/
The text was updated successfully, but these errors were encountered:
So here is the thing. When we do metacal on our simulations as an input to the network, we can either apply metacal on existing images (i.e. which already contain a psf and noise) like in normal metacal, or we can compute the response to a "true shear" introduced in the image before the noise and the PSF.
If we compute the response on existing images, we have to deal with the induced noise correlations that our shearing is doing. And for large shears it will be pretty significant. But, it means that we can use true images from the survey to train the network.
If we compute the response to a true shear, then we don't need to deal with the noise correlations and such. But we can only do this on simulations, because the need to shear the image before applying noise and psf, so not on the actual survey images.
The clean thing we might want to do is this second option, at the cost of not being able to directly use survey images :-/
The text was updated successfully, but these errors were encountered: