48 panel mosaic settings?

Hi Kees,

I will release a new version today which will have some adjustments in the star analyser module. This will enable you to lower the 

"detect above noise kappa" to 1

and 

"clip star profile" to as low as 0.01 (instead of 0.1)

Both settings should help in detection of more faint stars which we need to get a more robust result. So this could already have a positive effect on the end result.

Is the setting of "same camera and optics"off giving better results than "on"? 

Regarding improved mosaic algorithms: they will come but the main priority currently is the documentation and video tutorials to get everyone started with APP.

I think I can have a first redesigned mosaic algorithm ready in a couple of weeks (which we can test then with your data), which would be more robust for such large mosaics, to answer your question.

Thank you for posting the current entire mosaic, it looks very impressive 😉 !

Kind regards,

Mabula

Posted by: Mabula Haverkamp

 

Is the setting of "same camera and optics"off giving better results than "on"? 

I don't know, it is always off, have never tested this as i have no idea how to use this function......

Posted by: Kees Scherer

Posted by: Mabula Haverkamp

 

Is the setting of "same camera and optics"off giving better results than "on"? 

I don't know, it is always off, have never tested this as i have no idea how to use this function......

 

I you enable it, the algorithms work differently for both 2-view and multiple view registration.

If you load in (calibrated) light frames, not affected by a registration process earlier, the result with "same camera and optics" enabled is in most cases better and more robust.

Please try it and let me know what happens 😉

Mabula

I'm interested as to the # of overlapping frames setting... I'm guessing that it doesn't refer to the number of panes in your mosaic, which is what the wording suggests to me..... 

What does it mean and what does it achieve?

Posted by: Sara

I'm interested as to the # of overlapping frames setting... I'm guessing that it doesn't refer to the number of panes in your mosaic, which is what the wording suggests to me..... 

What does it mean and what does it achieve?

Hi Sara,

This setting only has influence when you are using:

1. Mosaic registration mode
2. Enabled Dynamic Distortion Correction
3. Disabled "same camera and optics"

What it does: this setting influences the distortion model per panel of the mosaic. Each panel will have different distortion model parameters.

If you set this value higher the model should become more robust. In a lot of cases , a value of 3 is fine especially for mosaics smaller than 3x3 panels, (so 2x3, 3x2, 2x2, 1x2, 2x1)

For bigger mosaics you will need to increase this value to get better optical distortion correction if you are running with the above three settings.

This parameter is part of the current: mosaic_not_same_camera_and_optics distortion_model calculations.

(I will upgrade these mosaic models soon and this parameter won't be needed then due to better formulation of the problem, so this parameter should become deprecated I think)

I found this cryptic message this morning. After clicking ok APP continues with integration, but stops after a while with the out of memory message.. It seems like APP is not using any SWAP space in Linux Mint as i have a 32GB swap partition in addition to the 2000 GB space for the APPwork directory and the 14 GB RAM allocation. There is no indication about the amount of "Memory" that i would need to complete the task?

Hi Kees,

Thank you for sharing, you are clearly running into the current limitations that APP has for image dimensions using the amount of RAM memory in your system.

In this case the limitation is the amount of RAM memory that APP can use, so this is not a software limitation but a hardware resource limitation.

See section 2.6 in the Quick Reference Guide ver 0.4 table 2.

your 14Gbs of memory can't handle more than the reported 1127 Mega Pixels that the integration field of view is composed of.

Apparently, the proposed downscaling of the integration so it still fits into your system did not work like it should. It reports the value of 0.0 as downscale factor.

Are you using APP 1.042 ?

Hopefully you haven't closed the APP session yet? Try to manually set the integration scale to a lower value than you had to be able to still do the integration.

can  you share how big the image would be which you are trying to create? Is it larger than the previous 48 panels?

Regarding swapping: if I were to allow APP to use more memory than the amount of available System RAM, than your OS will start to swap continuously. I would be happy to provide you with a version that can do this to test this, although I do think it will take a vey long time to complete then and your OS will probably freeze to death. It's something that isn't tested yet. Is the swap partition on a SSD drive? (in that case, it might not be too bad)

( I thought you mentioned earlier that the swap partition does not work in Mint 17.1 due to a Debian kernel problem? )

Mabula

Mabula, i am working on the 48 panel mosaic with 3 lights per panel. Each light is 16 megapixels, so the maximum size would be 768 megapixels, and with the 15% overlap around 600 megapixels, so i don't understand why APP 1.042 reports the 1127.43 megapixels even more strange with the 0.5 image scale (See previous screenprint). Swap on Linux Mint 17.1 can be activated manually with the swapon command. (On my other laptop i now run Linux Mint 18.1 XFCE, swap is working normally there and i also have a large swap configured. It is SSD)  I would like to test  a version supporting swap. I have not closed the session yet and will start a new integration run with image scale 0.3.

Hi Kees,

Thank you, okay.

What does the registration RMS column show? Can you show a screenprint with all the analytical results of the 48 panels by maximizing the frame list panel?

Okay, I am preparing the 1.043 build, I'll make a separate one for you where you can increase the memory amount of APP above the amount of system RAM.

Mabula

The 1127 megapixel memory problem message also pops up at image scales 0.3 and 0.1.

What does the registration RMS column show? Can you show a screenprint with all the analytical results of the 48 panels by maximizing the frame list panel?

I suspect an unstable registration making an Field Of View that extends to infinity.... that would explain the scale factor adjustment of 0.0.

You are right Mabula, the RMS values are all over the place, should have checked. But previous runs all showed RMS values around 0.1 strange enough. There are 48 panels, but 147 light frames so i include just 1 screenprint to show the effect. Have turned off same camera and optics now as that was the only change in settings.

Hi Kees,

Yes indeed, the mosaic registration failed, the parameters that were calculated are warping some frames extremely probably resulting in a gigantic Field Of View.

Let me know if it does give a stable result with the "same camera and optics" setting disabled. It will take quite a while for the algorithm to finish I think with 147 light frames with the current mosaic formulation in the algorithms.

Mabula

I did some extra testing with 6 panels and a total of 20 lights all loaded at once and processed into a mosaic. (5 panels with 3 lights and 1 panel with 5 lights). I now have a perfect result on scale 1.0 . See screenprint of a 100% sized detail where you can see an overlap and the effect of dithering. I used the following settings:

Mosaic settings APP:

3) Analyse stars
min size: 4
Clip Prof: 0.1
Filter profile on
Kappa: 4
Stars: 2000

4) Register

Pattern recognition: Quadrilaterals
scale start:  1
scale stop:  5
use dynamic distortion corr: on
same camera and optics: on
# of overlapping frames: 3
registration mode: mosaic
registration model: projective

6) Integrate

composition mode: full
LNC degree: 6
LNC it: 3
enable MBB: on
MBB%: 5
No under/overshoot on
Integrate set scale 1.0

Outlier rejection met 3 frames:
sigma clip, kappa=1.5, it=3  (low kappa value to eliminate bad columns)

Hi Kees,

Okay that's excellent to hear, maybe expand this test with 12 panels and the same settings.

I am surprised to see that scale 5 is enough here since you roughly have only 15 % overlap. But that's a positive surprise 😉

Mabula

Posted by: Mabula Haverkamp

Hi Kees,

I will release a new version today which will have some adjustments in the star analyser module. This will enable you to lower the 

"detect above noise kappa" to 1

and 

"clip star profile" to as low as 0.01 (instead of 0.1)

Both settings should help in detection of more faint stars which we need to get a more robust result. So this could already have a positive effect on the end result.

 

Kind regards,

Mabula

Hi Mabula, i have not seen the new version,  i assume that would be 1.043?

Yes, my apologies, there was some delay unfortunately. I'll upload 1.043 within a couple of hours 😉