StarXTerminator linear state

Euripides

Hello Adam, 

After viewing your latest video about Leonard, I came up again against an issue that I have already mentioned to Russell, but I would like share also here. I would say I'm pretty confident that you may have an idea about that.

So lets take the Leonard case scenario, with an OSC.

I run WBPP-->Calibrate-->Register and got output of R,G,B.

Run NSG on each channel.

So far so good. 

Now when I run StarXTerminator, while the starless version seems reasonable, the stars only images in every channel looks awful.

And of course, if I combine the channels, the result is even worse.

Any thoughts ?

PS : With non-linear data it works like a charm. 

BlockHead

Those are precisely what my stars look like as well.

(Did you go through my tutorial?)

You are just looking at them with an inappropriate stretch. 

(which I mention in the video, just adjust the screen stretch)

Keep in mind that the extraction of stars does not change the information meaningfully.

If you take that image, that you feel is crappy, and add it back to the starless image - you get your full image back. This is the second time this question has been asked of me today! I think I need to make a lesson on how neural nets work! 

-the Blockhead

BlockHead

Let me add, on the non-linear it WILL "work" better, since the definition between star and sky is easier. The neural net algorithm has been trained for this level of contrast.

-the Blockhead

Euripides

Hello Adam, 

I've followed the YouTube version, I am only a Fundamentals owner at the moment. I can recall that you've mentioned the stretch for the comet only image, I will recheck it for sure and give a try for a manual STF.

We would be more than happy to have this lesson too! 

Euripides

Ok I've just found the specific part but I could say that I am not capable to achieve a good result. 

Maybe its just an illusion or the way I see it, but those are not stars :-)

I can't recall that you have a video about this, starless - stars only linear stretch - if we need to apply anything "special", am I correct Adam?

BlockHead

I do not have a video on linear starless images. This is actually the first time I have used it in a meaningful way. Because the result is linear, you don't want to do any processing to the extracted stars (no processing that changes brightness values) before adding back or blending it with another image.

-the Blockhead

BlockHead

There is something else.....

There are two AI versions of StarXTerminator. 

The AI version 5 will not insert "noise" to match the background. This will make the extracted stars, even in linear images, probably appear better. However... even with AI 6, that low level noise matches your back ground.... so it should cause an issue.

-the Blockhead