PixInsight 1.8.7 Released

Juan Conejero said:
Hi Jeff,

I'm having an issue when using the Batch Preprocessing Script in the new 1.8.7.  The screen goes black after a few seconds and never recovers.

We cannot reproduce this problem. BPP runs normally on all of our working and testing machines, on all platforms.

So my question, is there a way to force PI to use the GeForce card and hopefully avoid the rendering issue?

On Windows, the PixInsight core application already attempts to use the discrete GPU exclusively on machines with dual graphics. This should work for both nVidia and AMD graphics cards. However, whether or not this works on a particular machine is a different thing; one never knows for sure on Windows, especially on laptops, where the concept of standards is pretty fuzzy. There is nothing more I can do to fix these issues.

What is really strange is that it works with 1.8.6 and not with 1.8.7. There are no changes at all that could justify such a different, and problematic, behavior.

I think this may be a memory issue.  I reinstalled 1.8.7 and re-ran the same file set.  100 bias, 100 darks, 60 flats, 60 lights.
The process failed when it began to integrate the bias frames, but did not crash the computer so I was able to get the log file.  The last portion is this:

2019-10-07 13:59:59] Reading FITS image: 16-bit integers, 1 channel(s), 3336x2496 pixels: done
[2019-10-07 13:59:59] Computing image statistics: done
[2019-10-07 13:59:59] Weight          :    1.00000
[2019-10-07 13:59:59]
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] Integration of 100 images:
[2019-10-07 13:59:59] Pixel combination .................. average
[2019-10-07 13:59:59] Output normalization ............... none
[2019-10-07 13:59:59] Weighting mode ..................... don't care
[2019-10-07 13:59:59] Scale estimator .................... MAD
[2019-10-07 13:59:59] Pixel rejection .................... Winsorized sigma clipping
[2019-10-07 13:59:59] Rejection normalization ............ none
[2019-10-07 13:59:59] Rejection clippings ................ low=yes high=yes
[2019-10-07 13:59:59] Rejection parameters ............... sigma_low=4.000 sigma_high=3.000 cutoff=5.000
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] * Available physical memory: 13.126 GiB
[2019-10-07 13:59:59] * Using 2496 concurrent pixel stack(s) = 12.438 GiB
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] Integrating pixel rows:    0 ->  2495:  80%
[2019-10-07 14:00:36] *** PCL Win32 System Exception: At address 00007FFE1E3EA839 with exception code C0000005 :
[2019-10-07 14:00:36] Access violation: invalid memory read operation at address 0000000000000000
[2019-10-07 14:00:36]
[2019-10-07 14:00:36] ************************************************************
[2019-10-07 14:00:36] * End integration of bias frames
[2019-10-07 14:00:36] ************************************************************

I read about disabling the automatic buffer size, so I did that in the ImageIntegration process (still failed, but gracefully), but is that setting also used in BPP?  Is there somewhere else I can check?

Thanks,
Jeff

 
Juan Conejero said:
...
The new automatic buffer sizes parameter of ImageIntegration is enabled by default. It is starting to become obvious that this has been a mistake, which I'll fix as soon as possible.
...
Actually, I would love if you kept that as a default, but consider available physical RAM when setting up the defaults. It is not trivial for the average user to decide on good values.

Georg


Georg
 
Juan Conejero said:
Hi Barry,

As I said in my answer to Brad's post, I cannot reproduce this problem. Version 1.8.7 works perfectly fine on all of our working Mac machines. All of them run macOS 10.14.6.

The only thing I can figure out here is a third-party application interfering with PixInsight. Do you have some virus protection application, or a similar utility?

Thanks Juan.

I do use Avast anti-virus however this has not changed since 1.8.6.  I have a late 2013 iMac with dual core i5, 16 Gb ram and 1Gb Nvidia GeForce video card.

Observations from testing and some curious behaviour of the real time preview of the HT, seemingly related to a project.

When I open a .xisf file on its own, ie the only file open within PI, the real time preview of the HT tool works very quickly as expected.  Great.

When I open a saved project, the real time preview of the HT tool is very slow (30 secs plus) to open.  I then moved one of the .xisf image windows into workspace 2 from workspace 1 (all of my windows and processes are in workspace 1) and this time the real time preview would not work at all, the circular button turned orange but no real time preview window appeared at all.  I then shut down the HT tool.  Upon re-opening the HT tool, I was able to use the real time preview on the .xisf image in workspace 2 and it worked quickly as expected behaviour.

I then moved the .xisf image back to workspace 1 and the real time preview window worked quickly, as expected behaviour.  I then saved the project and shutdown PI.

I then re-opened the same project by double clicking the project icon in Finder.  This time the real time preview worked quickly, as expected.

Not sure what to conclude from this - the project was started and saved after the update.  I have no other projects, either created with the new update or from an older version, to test.  However closing and re-opening the HT tool has seemed to 'cure' the slow behaviour.

HTH.

Barry
 
Barry-Wilson said:
Juan Conejero said:
Hi Barry,

As I said in my answer to Brad's post, I cannot reproduce this problem. Version 1.8.7 works perfectly fine on all of our working Mac machines. All of them run macOS 10.14.6.

The only thing I can figure out here is a third-party application interfering with PixInsight. Do you have some virus protection application, or a similar utility?

Thanks Juan.

I do use Avast anti-virus however this has not changed since 1.8.6.  I have a late 2013 iMac with dual core i5, 16 Gb ram and 1Gb Nvidia GeForce video card.

Observations from testing and some curious behaviour of the real time preview of the HT, seemingly related to a project.

When I open a .xisf file on its own, ie the only file open within PI, the real time preview of the HT tool works very quickly as expected.  Great.

When I open a saved project, the real time preview of the HT tool is very slow (30 secs plus) to open.  I then moved one of the .xisf image windows into workspace 2 from workspace 1 (all of my windows and processes are in workspace 1) and this time the real time preview would not work at all, the circular button turned orange but no real time preview window appeared at all.  I then shut down the HT tool.  Upon re-opening the HT tool, I was able to use the real time preview on the .xisf image in workspace 2 and it worked quickly as expected behaviour.

I then moved the .xisf image back to workspace 1 and the real time preview window worked quickly, as expected behaviour.  I then saved the project and shutdown PI.

I then re-opened the same project by double clicking the project icon in Finder.  This time the real time preview worked quickly, as expected.

Not sure what to conclude from this - the project was started and saved after the update.  I have no other projects, either created with the new update or from an older version, to test.  However closing and re-opening the HT tool has seemed to 'cure' the slow behaviour.

HTH.

Barry

Interesting, Barry. I have many projects saved with 1.8.6.xxx so I will verify your findings and report back to the thread.
 
Hi Juan,

Barry was correct. Previous projects saved in 1.8.6 that are loaded in 1.8.7 exhibit this issue. Good news is that saving the project under 1.8.7 seems to fix the issue. Here is a video displaying the effects in 1.8.6 and  before saving and after saving in 1.8.7

DropBox
 
Brad and Barry,

I am glad to know you have find a way to solve your issues with version 1.8.7 of PixInsight. However, we cannot reproduce these problems on any platform, on none of our machines. We can open projects generated with versions 1.8.4, 1.8.5 and 1.8.6 without any problems (and we have *really* complex test projects), and the HT tool works perfectly fine in all cases for real-time preview and view execution, on FreeBSD, Linux, macOS, and Windows.

Your solution is weird because nothing at all has changed in the HT tool from version 1.8.6 to 1.8.7. Nothing has changed in the real-time preview. Nothing essential has changed in projects, either, and the slight changes in 1.8.7 with respect to 1.8.6 have nothing to do with HT or R-T. However, if your solution works for you, who am I to dare question experimental evidence? :)

That said, if you can upload a project where this can be reproduced, that would be great for testing purposes. If I can reproduce the same problems you are experiencing, then I can understand what happens and find a solution.
 
Juan Conejero said:
Brad and Barry,

I am glad to know you have find a way to solve your issues with version 1.8.7 of PixInsight. However, we cannot reproduce these problems on any platform, on none of our machines. We can open projects generated with versions 1.8.4, 1.8.5 and 1.8.6 without any problems (and we have *really* complex test projects), and the HT tool works perfectly fine in all cases for real-time preview and view execution, on FreeBSD, Linux, macOS, and Windows.

Your solution is weird because nothing at all has changed in the HT tool from version 1.8.6 to 1.8.7. Nothing has changed in the real-time preview. Nothing essential has changed in projects, either, and the slight changes in 1.8.7 with respect to 1.8.6 have nothing to do with HT or R-T. However, if your solution works for you, who am I to dare question experimental evidence? :)

That said, if you can upload a project where this can be reproduced, that would be great for testing purposes. If I can reproduce the same problems you are experiencing, then I can understand what happens and find a solution.

Thanks, Juan. Understood and agree it's weird. When I get home tonight I will drop a 1.8.project on my dropbox. What do I need to ensure is there? The project .xosm and project.data?
 
Hi Brad,

Thank you so much. Yes, the .xosm file and the associated .data folder, preferably compressed as a zip file to prevent corruption. Or, if the project is bundled, the .pxiproject bundle, also compressed for the same reason.
 
Andres.Pozo said:
ario said:
I put 4 fotos 1-4 in
https://we.tl/t-e6anuf8IZO
1 is the right bottom corner of the mosaic.
I hope the link works.
Armando
Thanks ario. I can now reproduce the failure. However, it fails due to changes I am not responsible for. I have reported this so Juan is aware of it.

Thank you Armando and Andr?s. I have fixed this little (but annoying!) bug with an update I've just released for all platforms. Thank you for your patience and support.
 
Juan Conejero said:
Hi Brad,

Thank you so much. Yes, the .xosm file and the associated .data folder, preferably compressed as a zip file to prevent corruption. Or, if the project is bundled, the .pxiproject bundle, also compressed for the same reason.


Hi Juan,

DropBox

Same link, two new entries. 1) Compressed project data 2) another video showing another behavior. When I was testing and I clicked the real time preview I went to do some other things on the Mac. When I clicked on another app, taking focus away from PI, the real time preview instantly displayed. The video shows this as an example.

Let me know if you have any issues with the data - it's quite large.
 
Hi Brad,

I have tried to download this file several times without success. The file is big and I don't have access to our working network connection where I am now. I'll have to wait until next Tuesday, when I'll be back at our lab. Sorry for the inconvenience.
 
Juan Conejero said:
Hi Brad,

I have tried to download this file several times without success. The file is big and I don't have access to our working network connection where I am now. I'll have to wait until next Tuesday, when I'll be back at our lab. Sorry for the inconvenience.

No problem, Juan. It will be there so no worries
 
JeffSeivert said:
Juan Conejero said:
Hi Jeff,

I'm having an issue when using the Batch Preprocessing Script in the new 1.8.7.  The screen goes black after a few seconds and never recovers.

We cannot reproduce this problem. BPP runs normally on all of our working and testing machines, on all platforms.

So my question, is there a way to force PI to use the GeForce card and hopefully avoid the rendering issue?

On Windows, the PixInsight core application already attempts to use the discrete GPU exclusively on machines with dual graphics. This should work for both nVidia and AMD graphics cards. However, whether or not this works on a particular machine is a different thing; one never knows for sure on Windows, especially on laptops, where the concept of standards is pretty fuzzy. There is nothing more I can do to fix these issues.

What is really strange is that it works with 1.8.6 and not with 1.8.7. There are no changes at all that could justify such a different, and problematic, behavior.

I think this may be a memory issue.  I reinstalled 1.8.7 and re-ran the same file set.  100 bias, 100 darks, 60 flats, 60 lights.
The process failed when it began to integrate the bias frames, but did not crash the computer so I was able to get the log file.  The last portion is this:

2019-10-07 13:59:59] Reading FITS image: 16-bit integers, 1 channel(s), 3336x2496 pixels: done
[2019-10-07 13:59:59] Computing image statistics: done
[2019-10-07 13:59:59] Weight          :    1.00000
[2019-10-07 13:59:59]
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] Integration of 100 images:
[2019-10-07 13:59:59] Pixel combination .................. average
[2019-10-07 13:59:59] Output normalization ............... none
[2019-10-07 13:59:59] Weighting mode ..................... don't care
[2019-10-07 13:59:59] Scale estimator .................... MAD
[2019-10-07 13:59:59] Pixel rejection .................... Winsorized sigma clipping
[2019-10-07 13:59:59] Rejection normalization ............ none
[2019-10-07 13:59:59] Rejection clippings ................ low=yes high=yes
[2019-10-07 13:59:59] Rejection parameters ............... sigma_low=4.000 sigma_high=3.000 cutoff=5.000
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] * Available physical memory: 13.126 GiB
[2019-10-07 13:59:59] * Using 2496 concurrent pixel stack(s) = 12.438 GiB
[2019-10-07 13:59:59]
[2019-10-07 13:59:59] Integrating pixel rows:    0 ->  2495:  80%
[2019-10-07 14:00:36] *** PCL Win32 System Exception: At address 00007FFE1E3EA839 with exception code C0000005 :
[2019-10-07 14:00:36] Access violation: invalid memory read operation at address 0000000000000000
[2019-10-07 14:00:36]
[2019-10-07 14:00:36] ************************************************************
[2019-10-07 14:00:36] * End integration of bias frames
[2019-10-07 14:00:36] ************************************************************

I read about disabling the automatic buffer size, so I did that in the ImageIntegration process (still failed, but gracefully), but is that setting also used in BPP?  Is there somewhere else I can check?

Thanks,
Jeff

With the bug fix that was just released, BPP ran flawlessly on 100 bias, 100 darks, 60 lights and 80 flats.
Looking at the system information it looks like the process was limited to about 80% of the available RAM.

Thanks!!

Jeff
 
Is it intentional that there is no splash screen in Windows 10 any more?  Or is that a symptom of having only Intel graphics?
 
The splash screen works perfectly in version 1.8.7 on all platforms, and is enabled by default as always. Check if you have disabled it with Edit > Global Preferences > Main Window / Startup > Show splash at startup.
 
Brad said:
Juan Conejero said:
Hi Brad,

Thank you so much. Yes, the .xosm file and the associated .data folder, preferably compressed as a zip file to prevent corruption. Or, if the project is bundled, the .pxiproject bundle, also compressed for the same reason.


Hi Juan,

DropBox

Same link, two new entries. 1) Compressed project data 2) another video showing another behavior. When I was testing and I clicked the real time preview I went to do some other things on the Mac. When I clicked on another app, taking focus away from PI, the real time preview instantly displayed. The video shows this as an example.

Let me know if you have any issues with the data - it's quite large.

Hi Brad,

Thank you for uploading this project. It has helped me to reproduce and understand the problem. This is a complex bug reproducible on all platforms. I am working on it right now, so it will be fixed in the next update. Thank you for your patience and sorry for the trouble.
 
Juan Conejero said:
Brad said:
Juan Conejero said:
Hi Brad,

Thank you so much. Yes, the .xosm file and the associated .data folder, preferably compressed as a zip file to prevent corruption. Or, if the project is bundled, the .pxiproject bundle, also compressed for the same reason.


Hi Juan,

DropBox

Same link, two new entries. 1) Compressed project data 2) another video showing another behavior. When I was testing and I clicked the real time preview I went to do some other things on the Mac. When I clicked on another app, taking focus away from PI, the real time preview instantly displayed. The video shows this as an example.

Let me know if you have any issues with the data - it's quite large.

Hi Brad,

Thank you for uploading this project. It has helped me to reproduce and understand the problem. This is a complex bug reproducible on all platforms. I am working on it right now, so it will be fixed in the next update. Thank you for your patience and sorry for the trouble.

Thanks, Juan! Glad you were able to locate it and good luck and look forward to the fix!!!!
 
Just a follow up on this. This bug is now fixed in version 1.8.8 of PixInsight, which I'll release very soon. As an added benefit, the real-time preview window is now faster and more responsive to tool parameter changes. There are other planned improvements to this interface that I'm considering to implement also in the new version, not that I have opened this 'can of worms' :)

Thank you again for your help and support.
 
Juan Conejero said:
Just a follow up on this. This bug is now fixed in version 1.8.8 of PixInsight, which I'll release very soon. As an added benefit, the real-time preview window is now faster and more responsive to tool parameter changes. There are other planned improvements to this interface that I'm considering to implement also in the new version, not that I have opened this 'can of worms' :)

Thank you again for your help and support.

Thanks, Juan! You're awesome!!!!
 
Back
Top