Peculiar Number of Flagged 1-D Gaussians

[AKHughes1994]

Hello,

I'm trying to perform source extraction on a particular image (using versions 1.10.2), however, my source catalog has a large number of components with 0.0's for the image plane major and minor axis, which also causes incorrect values for the fluxes/positions.

Moreover, when I set fix_to_beam=True the fraction of flagged Gaussians increases and I get this error:
WARNING: After deconvolution, more than 50% of Gaussians are 1-D. Unless you're fitting an extended source, the beam may be incorrect.

The code outputs the correct beam shape (1.96136e-03, 1.47527e-03, 160.0 degrees) and it is not 1-D. Since we are fixing the Gaussian components to be the shape of the beam, and the beam isn't 1-D, I'm assuming it should be impossible to produce this error. Plotting the fits also shows well-fitting, and properly shaped Gaussian components. I've included an image with one of the sources that are zeroed in my output catalog but appear fine in show_fit().

I'm not sure what is going on. Here is a link (https://www.dropbox.com/s/rt1349e4eig5am7/img_1684711875_sdp_l0_1024ch_SAXJ1810.ms_pcalmask-MFS-image.fits?dl=0) to the image I'm trying to fit. Let me know if you need any more information or if you have already run into this issue in the past. The PyBDSF run is as follows:

img = bdsf.process_image(image_name,thresh_pix=5.0, fix_to_beam=True, minpix_isl=5.0, rms_box=(90,30), trim_box=(6750,7250,6750,7250))

img.write_catalog(format='ascii', catalog_type='srl', outfile = 'temp.txt', clobber=True)
    
img.show_fit(ch0_flagged=True)

[darafferty]

Thanks for the detailed report (and supplying the image!). I will take a look and see if there is something going wrong.

Just to be clear: do you see these 1-D Gaussians in the undeconvolved values (i.e. Maj and Min columns, see https://pybdsf.readthedocs.io/en/latest/write_catalog.html#output-cols) or only in the deconvolved ones (DC_Maj and DC_Min)? If only the latter values have zeros, then I think it is probably fine and you can ignore the warning, as the deconvolution algorithm is not terribly dependable, given the presence of noise and the fact that the beam and the PSF may differ somewhat. (In other words, most of the 1-D Gaussians are likely unresolved sources, but to the deconvolution algorithm they look slightly resolved along one axis)