Ha. Well, I suspect the larger astrophotography community is looking for something with a little more entertainment value. Thanks for watching this boring channel anyway, Richard!
Loved this data analysis, it's really refreshing to see content from people that knows what they are talking about and can present meaningful results. Thanks for the great work, keep it up! :)
Very, very interesting analysis, thank you so much for taking the time to do it! It must have been quite a bit of work. I personally have my weight almost completely reliant on number of stars, as higher FWHM results in fewer stars detected (as you suggested in one of your AF videos) - the big reason for doing so is my light pollution blanket: any time the target dips towards the horizon, contrast suffers (even though PI's SNR value actually increases). Fortunately # of stars helps me reject and weight that. I should be trying a similar experiment at some stage!
Thanks for watching, Cuiv. I think using the number of stars is a stars great metric, particularly for shorter focal length systems (many many stars). What is you experience using that metric with your SCT? I'm only getting 50 to 70 stars which would seem to make that metric a bit coarse for long focal length optical trains. Thoughts?
@@Aero19612 Good point - when I had my SCT, I wasn't very sophisticated and was visually rejecting frames (fewer frames too), so I can't really compare. I also have the "advantage" of city smog and LP really having a large impact the closer the target gets to the horizon, so for me even 50-70 stars would likely give me a "nice" descending curve, like I see now with my Newt.
As it indirectly relates to this topic... would you target total integration time per stacked sub/filter, total subs overall, or another variable? How do you feel the exposure time impacts final outcomes? Of course, there are so many technical variables (gain, image train, etc) and uncontrollable variables (seeing conditions, Bortle, etc). So many of us take images on 30s boundaries (30/60/90/.../300/...) but what if we could shorten our individual subs by 33% and go from 30s to 20s, etc? More total subs each with less impact from guiding, satellites, etc. Or maybe we use the time we get back for more frequent dithering without taking away from available imaging time. Of course, now we need to think about how much light we capture per sub when taking shorter exposures... plus the requirement to create additional Master Calibration frames for the exposure times. My guess is that ‘it depends’. :) Thoughts? Do you have a total integration time in mind when you select an imaging target? Do you have a method for selecting exposure times?
I generally think in terms of total integration time. In theory, total integration time is the key assuming you're not over-impacted by read noise at very short exposure lengths. Now, of course, the image quality varies from image to image, which is why I did this video. If image quality were the same for all images, you would see the SNR steadily improve with total integration time (regardless of individual frame exposure time), until the "square root" bites back and you start seeing the SNR improvement rate slow down until it's just not worth continuing to accumulate exposure time. Presumably, FWHM and Number of Stars would remain constant (when all images have same quality). But, alas, image quality varies a good bit, particularly at long focal length, because of seeing, wind, moon, position in sky, etc.
Thanks, well worth time spent watching your video. I found that they're both good pictures, both have the same details but overall the one on the left is the best picture, it has more depth and more confortable for the eye to see.
Nice video, thanks for doing this. I am wondering about your conclusions though about a cutoff line with SNR. Your plot of SNR vs count indicates that there was actually a pretty small increase in the number of images in the last few sets of low quality images for all the targets. This should mean a pretty small increase in SNR for the low quality images, which your plots show. I think one needs to plot the change in SNR vs the change in image count to get a real understanding of the benefits/detriments. Thanks again.
Thanks for watching, Larry! I like the idea of plotting the derivative of the curves (if you're not familiar with the term "derivative", you just reinvented calculus!). That would show off the break in the curve more clearly, like a step function. That is true, a few really bad image in the SNR vs count curve create that bend. If I understand you correctly, I think my cut-off at 10ish is well below those last few images. If I only use images with a score < 10, I'm tossing out 28% (79 of 281) of the total luminance set for M51 whereas the bend seems to occur in the final 40 images. To be fair, though, I doubt I could discern a visible difference between excluding 28% vs 14% since the worst images are those final 40.
Hi, thanks for all the great content that you constantly put on your channel. You might have said it - and probably I missed it - but when you integrated those images did you use "weighted integration" (with the keyword from SubframeSelector) or "noise evaluation integration"? If you used weighted, it might explain why there is not much difference between stopping earlier or including more images, as the "worse" images give a lower contribution to the total integration (by the weight applied to them).
Endless-Sky I Agree James always has such excellent well thought out content. I also wondered if a weighted integration was used, in which case those poorer subframe may not have contributed much. Would be interesting to repeat with staking all images using the don't care option in pixinsight
These are good thoughts, guys! I used the same integration settings I would use for creating a stack I would use for processing (i.e., "Noise Evaluation"). I just ran a stack with "Don't Care = 1". Here's a comparison for same number of images in the stack (202) Noise Weighted Don't Care FWHM 8.249 8.252 # Stars 32 32 SNR 11.19 11.17 So, not much difference at all when using the Noise Evaluation as a weighting parameter. I haven't been adding the subframeselector weight to the FITs header, but I would it would have a greater effect. Thanks for watching!
I think the lesson I'm taking from this is it doesn't really *HURT* to add poorer quality images when it comes to what the eye can tell. By cutting out those images I might speed up processing time, but what I gain in processing time I lose in organizing them out of the stack time. So maybe I'll just keep 'em in for ease of process (most of the processing is done AFK anyway).
You're not wrong, Dennis! There seem to be a lot of these types of fuzzy trade-offs. Deconvolution, for example, takes a long time and I only see marginal improvement. Likewise here. I suppose, philosophically, I'll keep trying to do "the little things" in the hopes they'll add up to a "good thing". Whatever extra time I spend processing, it's a drop in the bucket to the amount of time I spend collecting the data in the first place. Thanks for watching!
watched vid. love it. lets see the comparison of the two just after stacking. Perhaps all the post process essentially eliminates any differences between the two.
That's an interesting find. I had been wondering if I could use a full stack with lower quality images just for the luminance channel. To add some extra snr to the final image. But I haven't tried that yet. Thanks for the video.
Thanks for watching, Nick! Always seems like hot-shot numbers are balanced by "what do you actually see". Same story here. There's some benefit, but it doesn't "kill" an image to include poorer quality images (of course, it's not helping either!)
I usually try to find a sudden cut-off in quality during stacking. And include everything above that. I'm still a bit of a newbee in processing though. So the image quality is not my biggest concern yet. ;)
It would be interesting to compare say the stacked image of the worst 66% to the best 33% and various other ratios, How many bad images does it take to beat a few good images.
Another great video. The problem I have is that when imaging a star poor field like M101 with our similar setup, if the quality of the subframes drops very much that DSS wont accept them for stacking at all and I'm forced to stack whatever I can. I'm rarely in a position that I can afford to manually throw away frames. Sometimes I look at a subframe in a raw Fits viewer and stars seem ok but there is some slight out of focus and I just cant get DSS to accept them. Any workarounds that you have tried to get DSS to find enough stars.
One BIG work-around for DSS: I don't use it! I use the calibration routines in PixInsight : ImageCalibration, StarAlignment, and ImageIntegration. In my early early days (before I had PixInsight), I tried DSS. I found it to be very temperamental (of course, it is free. So there's that). It's painful to throw out what looks like a good sub.
last night i shot 180 subs x 60 sec of IC-5146 with the big bad moon , now after watching this its going to take 6 hours of testing to find out what works.,...
Then my work here is done! Or, as Dennis commented below, "to heck with it" and just stack everything. Our eyes are far less sensitive than our quantitative metrics. I'm looking forward to getting some sky time to do the Cocoon Nebula. Should fit nicely in the C9.25 field of view. Thanks for watching, Mitch!
Thank you, James. You’ve answered this question for me rather convincingly. I’m very appreciative of your hard, detailed work.
Thanks, Scott. It's a murky area so feel free to experiment. I'm not sure there's a universal "right" answer. Clear Skies!
Fantastic video. Really appreciate the quantitative approach. Very informative. Thanks!
Thanks Ferdi! And thank you for watching!
James, thanks for sharing all of your well researched experimenting! You are awesome!
Thanks for watching, Elmiko! I hope you found it useful for your efforts.
Don't understand how you only have 2K subscribers, James. I hope you keep doing what you're doing because your videos are fantastic.
Ha. Well, I suspect the larger astrophotography community is looking for something with a little more entertainment value. Thanks for watching this boring channel anyway, Richard!
Loved this data analysis, it's really refreshing to see content from people that knows what they are talking about and can present meaningful results. Thanks for the great work, keep it up! :)
Thanks for watching! Who said I knew what I was talking about? Never stopped me from talking before.
fantastic analysis as always James! Everyone appreciates the time you put in this and give us great information. Thanks a lot
Hope it is useful to you. Thanks for watching, Matteo!
Very, very interesting analysis, thank you so much for taking the time to do it! It must have been quite a bit of work. I personally have my weight almost completely reliant on number of stars, as higher FWHM results in fewer stars detected (as you suggested in one of your AF videos) - the big reason for doing so is my light pollution blanket: any time the target dips towards the horizon, contrast suffers (even though PI's SNR value actually increases). Fortunately # of stars helps me reject and weight that. I should be trying a similar experiment at some stage!
Thanks for watching, Cuiv. I think using the number of stars is a stars great metric, particularly for shorter focal length systems (many many stars). What is you experience using that metric with your SCT? I'm only getting 50 to 70 stars which would seem to make that metric a bit coarse for long focal length optical trains. Thoughts?
@@Aero19612 Good point - when I had my SCT, I wasn't very sophisticated and was visually rejecting frames (fewer frames too), so I can't really compare. I also have the "advantage" of city smog and LP really having a large impact the closer the target gets to the horizon, so for me even 50-70 stars would likely give me a "nice" descending curve, like I see now with my Newt.
Nice analysis, that answered some questions I had been having. I also enjoyed your last four videos very much.
Thanks Pat! I think we all trip over the same questions. Hope this was useful!
Very well done! Thanks for your time and effort putting these types of videos together.
No problem, Jeff. Just trying to make my processing better and thought I'd share what I'm doing.
Really interesting, thanks for doing this 👍🏻
Thanks for watching, Olly!
As it indirectly relates to this topic... would you target total integration time per stacked sub/filter, total subs overall, or another variable? How do you feel the exposure time impacts final outcomes? Of course, there are so many technical variables (gain, image train, etc) and uncontrollable variables (seeing conditions, Bortle, etc). So many of us take images on 30s boundaries (30/60/90/.../300/...) but what if we could shorten our individual subs by 33% and go from 30s to 20s, etc? More total subs each with less impact from guiding, satellites, etc. Or maybe we use the time we get back for more frequent dithering without taking away from available imaging time. Of course, now we need to think about how much light we capture per sub when taking shorter exposures... plus the requirement to create additional Master Calibration frames for the exposure times. My guess is that ‘it depends’. :) Thoughts? Do you have a total integration time in mind when you select an imaging target? Do you have a method for selecting exposure times?
I generally think in terms of total integration time. In theory, total integration time is the key assuming you're not over-impacted by read noise at very short exposure lengths. Now, of course, the image quality varies from image to image, which is why I did this video. If image quality were the same for all images, you would see the SNR steadily improve with total integration time (regardless of individual frame exposure time), until the "square root" bites back and you start seeing the SNR improvement rate slow down until it's just not worth continuing to accumulate exposure time. Presumably, FWHM and Number of Stars would remain constant (when all images have same quality). But, alas, image quality varies a good bit, particularly at long focal length, because of seeing, wind, moon, position in sky, etc.
Excellent thank you!
Thanks Tim! And thanks for watching!
Thanks, well worth time spent watching your video. I found that they're both good pictures, both have the same details but overall the one on the left is the best picture, it has more depth and more confortable for the eye to see.
Thanks for the feedback, Filipe! It is difficult to separate “detail” from “color”. I’ll take another shot at the processing and hopefully do better.
@@Aero19612 Maybe you can just convert both images to lightness so they can be better evaluated without regard to colour
@@reefsteve That's a very good idea, Steve. That would focus the eye on the detail without being distracted by the color. Thanks for watching!
So I'd be interested in knowing how you calculate that first number that you attach to each image - 0669, 0687 etc.
Nice video, thanks for doing this. I am wondering about your conclusions though about a cutoff line with SNR. Your plot of SNR vs count indicates that there was actually a pretty small increase in the number of images in the last few sets of low quality images for all the targets. This should mean a pretty small increase in SNR for the low quality images, which your plots show. I think one needs to plot the change in SNR vs the change in image count to get a real understanding of the benefits/detriments. Thanks again.
Thanks for watching, Larry! I like the idea of plotting the derivative of the curves (if you're not familiar with the term "derivative", you just reinvented calculus!). That would show off the break in the curve more clearly, like a step function. That is true, a few really bad image in the SNR vs count curve create that bend. If I understand you correctly, I think my cut-off at 10ish is well below those last few images. If I only use images with a score < 10, I'm tossing out 28% (79 of 281) of the total luminance set for M51 whereas the bend seems to occur in the final 40 images. To be fair, though, I doubt I could discern a visible difference between excluding 28% vs 14% since the worst images are those final 40.
Hi, thanks for all the great content that you constantly put on your channel.
You might have said it - and probably I missed it - but when you integrated those images did you use "weighted integration" (with the keyword from SubframeSelector) or "noise evaluation integration"? If you used weighted, it might explain why there is not much difference between stopping earlier or including more images, as the "worse" images give a lower contribution to the total integration (by the weight applied to them).
Endless-Sky I Agree James always has such excellent well thought out content. I also wondered if a weighted integration was used, in which case those poorer subframe may not have contributed much. Would be interesting to repeat with staking all images using the don't care option in pixinsight
These are good thoughts, guys! I used the same integration settings I would use for creating a stack I would use for processing (i.e., "Noise Evaluation"). I just ran a stack with "Don't Care = 1". Here's a comparison for same number of images in the stack (202)
Noise Weighted Don't Care
FWHM 8.249 8.252
# Stars 32 32
SNR 11.19 11.17
So, not much difference at all when using the Noise Evaluation as a weighting parameter. I haven't been adding the subframeselector weight to the FITs header, but I would it would have a greater effect. Thanks for watching!
I think the lesson I'm taking from this is it doesn't really *HURT* to add poorer quality images when it comes to what the eye can tell. By cutting out those images I might speed up processing time, but what I gain in processing time I lose in organizing them out of the stack time. So maybe I'll just keep 'em in for ease of process (most of the processing is done AFK anyway).
You're not wrong, Dennis! There seem to be a lot of these types of fuzzy trade-offs. Deconvolution, for example, takes a long time and I only see marginal improvement. Likewise here. I suppose, philosophically, I'll keep trying to do "the little things" in the hopes they'll add up to a "good thing". Whatever extra time I spend processing, it's a drop in the bucket to the amount of time I spend collecting the data in the first place. Thanks for watching!
watched vid. love it. lets see the comparison of the two just after stacking. Perhaps all the post process essentially eliminates any differences between the two.
Check out the vid at 11:05. I show a comparison of the two luminance images at that point. Thanks for watching!
That's an interesting find. I had been wondering if I could use a full stack with lower quality images just for the luminance channel. To add some extra snr to the final image.
But I haven't tried that yet.
Thanks for the video.
Thanks for watching, Nick! Always seems like hot-shot numbers are balanced by "what do you actually see". Same story here. There's some benefit, but it doesn't "kill" an image to include poorer quality images (of course, it's not helping either!)
I usually try to find a sudden cut-off in quality during stacking. And include everything above that.
I'm still a bit of a newbee in processing though. So the image quality is not my biggest concern yet. ;)
It would be interesting to compare say the stacked image of the worst 66% to the best 33% and various other ratios, How many bad images does it take to beat a few good images.
That would be an interesting comparison. Good suggestion, Mike. Thanks for watching! FWHM is in the eye of the astrophotographer.
Another great video. The problem I have is that when imaging a star poor field like M101 with our similar setup, if the quality of the subframes drops very much that DSS wont accept them for stacking at all and I'm forced to stack whatever I can. I'm rarely in a position that I can afford to manually throw away frames. Sometimes I look at a subframe in a raw Fits viewer and stars seem ok but there is some slight out of focus and I just cant get DSS to accept them. Any workarounds that you have tried to get DSS to find enough stars.
One BIG work-around for DSS: I don't use it! I use the calibration routines in PixInsight : ImageCalibration, StarAlignment, and ImageIntegration. In my early early days (before I had PixInsight), I tried DSS. I found it to be very temperamental (of course, it is free. So there's that). It's painful to throw out what looks like a good sub.
last night i shot 180 subs x 60 sec of IC-5146 with the big bad moon , now after watching this its going to take 6 hours of testing to find out what works.,...
Then my work here is done! Or, as Dennis commented below, "to heck with it" and just stack everything. Our eyes are far less sensitive than our quantitative metrics. I'm looking forward to getting some sky time to do the Cocoon Nebula. Should fit nicely in the C9.25 field of view. Thanks for watching, Mitch!