16.52
#...
·
 |
|---|
|
Hi al, Progress on 40mm southern survey. First pass at 17 fields; all fields excluding 4, 5, 11, 12, 13, 20, 21, 22, 23, 28 and 29. Should still manage to get 19, 28 and 29 later on this lunation, but only when the aurora subsides. I attach an image; this time created in Zenithal equal area projection using MosaicByCoordinates and GradientMergeMosaic. Compressed 2arcmin per pixel for transfer  |
3.94
#...
·
|
|---|
| @Brian Boyle I played with your data and found some interesting details on how to use MosaicByCoordinates , I will open one or two new topics soon and would love to see some feedback from you. Michael |
|
16.52
#...
·
·
1
like
|
|---|
| @Michael Ring thanks for all that material. I will review over the coming days - along with shimming my 40mm Sigma rig to see if I can get better images. Currently I achieve "best focus" with the barrel at about 2m. FWHMEccenticity say no tile, but images go off quite rapidly into the corners. I am thinking the increasing the back focus distance should increase the best-focus setting on the barrel to inf. [Disappoiinting really since the ZWO adapter is supposed to give best back focal distance for a Canon lens. In the meantime is a 2arcmim moscaic of all 21 (out of 29) fields imaged so far. As you can tell the ones nearest to the horizon will been to be re-imaged next lunation. At the rest completed later in the year. But overall I think it is a reasonable start. Pipeline was: WBPP -> GraXpert -> BXT(correct only) -> MosaicByCoordinates (Zenithal Equal Area Centre 13h -60) -> GraidentMergeMosaic).  |
|
3.94
#...
·
|
|---|
| @Brian Boyle , would you mind uploading the new stacks to your Dropbox? Weather forecast for the weekend is clouds and thunderstorms so I might waste some time on playing with your data again and will perhaps also add the fields I have, which will be a good test to see if what I think i learned about mosaicbycoordinates is actually true. Michael |
|
16.52
#...
·
·
1
like
|
|---|
Michael Ring: Sure thing - just trying to improve Fields 27 10 and 19 when the moon sets tonight. But we also have auroral activity, so we will see. Would like to try Field 1 again, but think the aurora will definitely rule rhe one out. Will put images in the DropBox when I am done. |
|
3.94
#...
·
|
|---|
|
Take your time, no worries.... I just checked the weather report 200km away and there we will likely have sunshine and clear nights so we will do a round of Astro-Camping this weekend..... Michael |
|
16.52
#...
·
·
1
like
|
|---|
|
Here is my latest effort, with 21/29 southern fields. Processing as before. Stretch just STF. 2arcmin resolution. I re-took four fields, as the biggest challenge is to match fields with differing transparencies. Quite hard to do particularly for fields that go to +15 and are only 15 degrees above the mountains from my place. Will now put all the WBPP files in dropbox for people to experiment with. They will be in the Widefield subfolder of the ABC Survey folder of the dropbox site. If you don't have access please send PM me. It would be great to see any other attempts to match. @Michael Ring do you have some Sigma Art frames in the North you could add to the Mosaic? Good to get a bit more of Orion - now gone for the next four months of so. @James Tickner do you reckon you could run your latest matching code on this data? [Thanks for the field centre, they worked well]. I have tried to fill the gap at Field 13. I have some data here - but need to work more on it as I was getting close o the horizon. It may not be useable.  |
|
3.94
#...
·
|
|---|
| @Brian Boyle , weather changed again so we stayed home and I will play with your data over the (long) and cloudy weekend, will add the fields I have. Orion is definitely there and also in ok quality. I will also upload my northern stacks to your directory. Michael |
1.20
#...
·
·
1
like
|
|---|
|
Apologies for the long time offline - a combination of travel and work commitments means that I'm only just catching up with things. @Brian Boyle Great to see the progress with the 40 mm survey. Are you able to share the stacked images on Dropbox (if you haven't already)? I think they will help my gradient issue discussed below. I've continued to work intermittently on the mosacing. I'm now pretty happy with the alignment and star colour corrections. Attempts to patch up the radial gradient noted previously by Brian for my south pole mosaic have been less successful. Whilst I managed to get rid of the gross gradient by placing some constraints on the field brightness around the perimeter of the image, this resulted in unsightly artefacts where fields overlap - kind of a 'waffle' effect. I think the iterative approach I'm using to match gradients between fields just doesn't converge reliably enough. I want to go back to an approach I described a new months ago, namely fitting a smooth gradient to a large area of sky (in principle, the whole sky once we have that much data) and then correcting individual fields to match this underlying gradient field. This is where the 40 mm low-res data will come in! These data would provide an excellent starting point for the underlying gradient estimation. A couple of other points: - I noted some comments about difficulty using the PI mosaic tools where image distortion or field curvature becomes a problem. If they would be useful, I'm happy to upload my rebinned images that have been astrometrically corrected back to a consistent projection. This should remove these issues and allow for PI's gradient matching tools to be used. Let me know if anyone's interested! - I see that we've now completed just over 3 octants of the sky (~38.5%) so I should release a new whole-sky image. |
|
3.94
#...
·
|
|---|
| @James Tickner , I have some low data fields that I could stack (usually less than an hour of data) if you are interested I can stack those, will make the north look not that empty... Ping me if you are interested to include them in the new whole-sky image Michael |
|
16.52
#...
·
·
1
like
|
|---|
| @James Tickner all images are in the wide field subdirectory of the ABC survey on DropBox. If anyone doesn't have access that wants it, please PM me. I have tried a few alternative pipelines. Here is what I have found. YMMV. 1) GraXpert + BXT (correct only) before MosaicByCoodinates seems to work OK. @Michael Ring I don't understand the difference between putting the central coordinates in the main menu [which I do] and putting them in the advanced tab. 2) Using SPCC before mosaicing doesn't give very good results 3) Scretching before mosaicing gives interesting results - bringing out faint stuff [i.e. Pyxis and Puppis Nebulae] but might also show up terrestrial cloud. 4) PTGui does a wonderful job mosaicing, but needs non-linear images. Stretching Milky Way panels [using the sretch PixelMath expressed invented by Juan] gives good results along the plane, but the out-of-plane frames don't match well. [The medians are the same, the non milky frame are too stretched.] Bottom line, every method has is advantages, but so far noting I have tried works 100%  |
|
16.52
#...
·
|
|---|
|
A further update. Here is a lightly stretched mosaic of the 21 southern 40mm fields. [GraXpert, BXT prior to mosaicing using MosaicByCoordinates and GradientMergeMosaic] Original done at 30arcmin/pix - rebinned to 2arcmin/pix for upload.  It's OK, but doesn't go very deep. If I push the stretch I get this  All the faint stuff is there [Puppis/Pyxis, Zeta Ooh nebulae, South Pole + Barnard's galaxy region IFN]. but so it this field-to-field variation left by GradientMergeMosaic. Has anyone had any better luck at removing this than me? Otherwise I might put out a general post to see what other solutions might be out there. |
|
1.20
#...
·
·
1
like
|
|---|
| @Brian Boyle Thanks for the update! This is pretty much exactly the problem I'm working on at the moment with our 'normal' ABC survey fields. The steps I'm following are: 1. Removal of gross gradients from individual fields (my own tool, broadly equivalent to GraXpert) 2. Rebinning based on astrometric solution 3. Adjustment of average RGB background levels for each field to the same value (typically 0.05, but doesn't matter too much). Note this has to be performed after the rebinning as a flux-preserving rebin can shift the background levels depending on the final pixel size 4. Matching gradients between fields <- The tricky bit! 5. Mosaicing with blending at field edges Looking at the highly stretched image you've shown above, I'm guessing the PI function isn't applying steps 3. or 4. Some of the fields including the Milkyway are just brighter overall. Oddly, you also seem to be getting increased brightness in the regions where two fields overlap, which would suggest that perhaps the blending function isn't normalising correctly? When I get a chance I'll download your files and have a play with my code. Regarding the gradient matching between fields, I'm looking to move away from my iterative matching approach - as noted previously, it seems to be just to finicky to control well. |
|
16.52
#...
·
|
|---|
|
Thanks @James Tickner - a great summary I have also posted to the main forum to see if anyone else has some tips/suggestions. I am a great believer in the wisdom of crowds approach. There is one rather obvious effect in the scretched image i.e. that the fields with the high backgrounds have the higher star counts [Milky Way fields] To me this suggests that the field-to-field background calculation is being biased by the large number of stars [higher counts overall] in the fields near the plane. There are enough stars to bias even robust estimators like the median when taken over the whole field. |
|
1.20
#...
·
·
1
like
|
|---|
| @Brian Boyle I think even median fails as a robust estimator when there's just too much going on in an image. I've seen some discussions of using a combination of median and mean (eg (1 + x) * median - x * mean, with 'x' dialed in to taste). I think the idea is that the median estimator is pulled up in a crowded field, and the mean estimator is pulled up more, so a weighted difference can provide some correction. But there doesn't seem to be a good prescription for finding 'x', so it all seems a bit hand-wavy. In my background estimations I did experiment with excluding a circular region sounding each star found in the plate-solving stage, but this didn't make too much difference. I'm looping back to the idea I posted a few months ago, namely dividing the complete mosaic image into tiles (maybe 200x200 pixels or similar) and estimating the median background in each tile. Where two or more fields overlap, the median background is calculated separately for each one, resulting in two or more estimates of background for that tile. A smooth 2D spline function is then fitted through all the estimated background values. For each field, a separate smooth 2D spline function is fitted through the background for just that field. Each field is then background corrected by the difference between its local background function and the global background function at the same location. In other words, the background levels for each field are 'dragged' to the globally smooth background. Not sure if that makes sense - it's hard to describe in words. It will be good to play with your 40 mm fields as memory and CPU requirements should be much less than for the full survey, making it easier to experiment. |
|
16.52
#...
·
·
2
likes
|
|---|
James Tickner: I do think the 40mm fields will be a good testing ground. I am going to try to repeat as many as I can in the next lunation, as I am going to improve the camera/lens connection with less vignetting and better back focal distance. Brian |
