Trying to use Eye

[npirzkal]

I would like to try using Eye to detect small knots in some images. It is not clear to me how one should set up the put mask images in order to train eye... Should these images contain just 0's and 1's? Should the entire feature be set to 1? or just the peak/center part of it?
Finally what should a proper mask end up looking like? And should the file produced by Eye simply substitute default.conv?
So far, all output from Eye result in Sextractor failing to detect anything in the same images that I used during the training of Eye...
Also, it seems that there is something wrong with Eye and FITS files. V. 1.3 does not seem happy with a single extension fits file and seems confused about the true number of extensions present in a fits file. Has anyone one got this problem?

[Emmanuel Bertin]

Dear Nor,

sorry for taking so much time to answer. Too many things to do.

Quote:Should these images contain just 0's and 1's?

No. But they should contain only null or positive pixel values.

Quote:Should the entire feature be set to 1? or just the peak/center part of it?

It is up to you. In general, one uses EyE to identify all the pixels that are believed to belong to some feature. But nothing prevents you from triggering only on the central pixel of some feature, as long as the latter is not too extended.

Quote:Finally what should a proper mask end up looking like?

What one typically does is to subtract the background from examples of the desired features, extract them and paste them on a blank image. In that way, the neural network learns to trigger only on pixels which belong to the features.

Quote:And should the file produced by Eye simply substitute default.conv

Yes!

Quote:Also, it seems that there is something wrong with Eye and FITS files. V. 1.3 does not seem happy with a single extension fits file and seems confused about the true number of extensions present in a fits file. Has anyone one got this problem?

Exact. I just noticed some problems while doing some tests with some MEFs.
Thanks for the notice.
I keep you posted. And once again, sorry for taking so much time to anwer.
- Emmanuel.