Support for CHI2 images and absolute thresholding?

[hjmcc]

Dear SE++ developers, thanks for all the hard work.

I have a question, is SE++ able to use CHI2 detection images produced by Swarp? In 'sextractor classic' these images were signalled by adding -THRESH_TYPE ABSOLUTE in the configuration file. I grepped the sources and documentation and couldn't find anything 😰😰😰 .

thanks !

[mkuemmel]

That parameter doe not exist, right. I am not aware we ever discussed this. I'll have a look. If it does not exist it shouldn't be difficult to implement.

Could you provide us with some example data (chi-square image plus SE2 catalog plus SE2 seg image)?

[marcschefer]

I'm not entirely sure what's in those images but we use the detection image to estimate the initial value of flux for model fitting so that could be a problem if doesn't give us a usable flux.

[mkuemmel]

That parameter to make the detection threshold absolute for CHI2 images does not exist yet, and we will implement it for the next version v0.17.

Until then you can:

• pair the CHI2 detection image with an rms-image all 1.0;
• set the background to 0.0;

That should result in the intended behaviour.

@hjmcc again, could you give us some example data?

[hjmcc]

Thank you @mkuemmel and @marcschefer for your help!

I put some images from COSMOS2020 (https://ui.adsabs.harvard.edu/abs/2021AAS...23721506W/abstract) in https://exchg.calet.org/hjmcc/sepp/

You can find a subimage of the chi-mean images in there, as well as the same segment of the i-band image. I include an image from the paper showing what the parameters we used in COSMOS2020.

My aim to is to try to run SE++ on the COSMOS2020 data with detection parameters as close as possible to SE2.

Thanks for the hint @mkuemmel , I will try that!

[mkuemmel]

@hjmcc I checked you images, the weight image is 1.0 everywhere, hence the detection threshold is absolute as expected for a chi-square image. This zip file: chi2.zip contains the configuration files (ascii and python) for the use case you described. Its kind of structural, so I did not try to put in all the parameters you listed above. It uses the chi-square image as detection and the i image as measurement image. There are some hints how to extend to several measurement images.

Have a look!

[hjmcc]

Thank you, @mkuemmel for the help and posting these scripts!

I tried your config and compared it to a similar config for SE2.0. Seems to work, it detects about ~20% more objects however, as you can see from the check-images (OBJECTS from SE2.0 and SEGMENTATION from SE++). So I understand that the SE++ deblending and splitting algorithm is different from SE2? I will make some more tests. GIF shows (chi2, SE2 and SE++). (We have quite an agressive threshold in SE2.0 to avoid too many spurious sources around the very bright objects in COSMOS.

[hjmcc]

For reference, this is what happens near a bright star. It's the same order: chi2, se2, se++. Some tuning still!

[mkuemmel]

Some comments from my side:

• In the first image the SE++ objects seem to be larger and the additional 20% objects are at the faint end. Maybe the min-area is different and/or the weighting of the RMS image (I used absolute weighting).
• The deblending algorithm is very similar to SE2. Note that starting from 0.16 grouping is on by default.
• Note that you can set a second core threshold and discard objects with a small number of pixels above that core threshold. This enlarges objects and suppresses false positives around bright extended objects.

On the second image I think I switched background subtraction off, which to me seems to be appropriate for a chi2 image.

[hjmcc]

Thanks again @mkuemmel. I confirm that the minimum area is the same in both cases.

Note that you can set a second core threshold and discard objects with a small number of pixels above that core >threshold. This enlarges objects and suppresses false positives around bright extended objects.

Thanks for the hint. I guess you are referring to these parameters:

The core threshold level

core-threshold-value=0

The minimum pixel area for partitioning

core-minimum-area=0

Activate core threshold partitioning

partition-corethreshold=0

What should I set them to? I couldn't find anything in the documentation...

Cheers!

[mkuemmel]

Set:

• core-threshold-value to your 'old' detection-threshold;
• core-minimum-area equal to detection-minimum-area;
• lower the value for detection-threshold;
• switch on the partitioning with partition-corethreshold=0

The objects get larger which eliminates the halo objects around bright galaxies. False positives in the field are suppressed by the 'original', high core threshold.