Dynamic Crop - you need to eliminate 'rough edges' around the periphery of the image
STF - to let you see what you are dealing with (noting that this doesn't actually 'change' the image in any way)
Since you define when you do STF, shouldn't you do it also before the dynamic crop?
Definitely crop should be the very first thing, then DBE.
A couple of times I have seen images from "superstars" that in the final JPEG had one of those rough edges due to dithering, misalignment or whatever.
And no, I won't name names. All I'd say is that the images were praised in the public forums despite the edge was clearly visible.
And I wouldn't go too quickly on the second DBE. I know Juan says: "you can apply DBE twice, that's ok". And yeah, it's ok, but I'd rather spend some time perfecting the 1st background model than going for a second pass. Sometimes I generate well over 5-7 background models before I nail it. And by perfecting I mean trying really hard to nail it, not just sampling more here or there. Plus the more you do it, the better you become at it.
Anyway, why do I try to aim for just one DBE if possible? Because, to me at least, the best way to know a DBE has been successful is by looking at the background model, even more than looking at the resulting image (both are important of course). If you're removing, say, a gradient, and the background model has slight variations that you simply know that can't be the shape of your gradient (even considering that the gradient may have been building up during the capture as your target moves, etc), your model is wrong, and you can try to correct it adjusting the DBE parameters and adding/removing samples. Once you apply the first DBE, your only guide is the final image, not the model - because you no longer will be correcting a gradient but the remains of having partially corrected a gradient, and so your 2nd model can look like anything. If you're imaging a pretty galaxy or nebula and don't care about the background, this is probably ok, but if you're after very faint background stuff, you can no longer trust you're removing crap and preserving the right signal.
I usually do 1 or 2 DBEs on RGB data (when I do LRGB, which is most of the time), but with L I try to stay with just 1 DBE. The only exception would be dealing with a gradient so cabrón that I can "read" clearly what hasn't corrected, so in the 2nd DBE I aim only for that uncorrected area (something on an edge or stuff like that)... I don't mind so much doing two DBE passes (if needed) with RGB for two reasons:
1) A RGB background model is not as easy to read, having technically three models in one (R, G and B). I could split it, but I don't go through that trouble in part because of reason #2:
2) It's "just" color, for my own stuff I'm often content with the fact that things get "colored" and my main goal is to see color gradients dissapear - faint signal isn't that important, for me, in the RGB data. If it was a "work for hire" I'd probably spend more time.
Anyway, this is how I approach this stuff and so far it seems to work well. Your mileage may vary.
Cheers,
Rogelio
