Starnet ++ with New Mac M1pro Chip, not working

anyone can write a module for pixinsight. your complaint is equivalent to asking why shareware of varying quality for windows or osx exists when you have to pay for the operating system - it makes no sense.

the original starnet was "adopted" by juan, and distributed with PI. this has not happened with starnet2.

starxterminator uses the PI distribution center so that installation/re-installation and maintenance of the module is very easy. starnet2 does not use this mechanism.

your actual beef is with how juan distributes updates to the 'core' of PI. when a new version is available, it has to be installed from scratch. this is what "resets" your installation with respect to scripts and modules. that's just how it is and as the author it's his choice in how to release new versions of the PI core.

rob
 
Folks - I am having trouble when I go to Modules/Install Modules on my MacBook Air M1. It says No Modules and when I try to open it up it and all other files in the bin folder are greyed out. Looks like I have some sort of permissions problem or something else. Any advice?
 
you can't open a module directly in PI. it has to be opened from the "install modules" menu. if it doesn't show up there, it means some dependency for the module has not been satisfied. in this case it is probably the tensorflow libraries. try quitting PI and moving the tensorflow libraries to

/Applications/PixInsight/PixInsight.app/Contents/Frameworks/

assuming you are using the libtensorflow* files that shipped with StarNetV2 and not the ones that ship with PI.

rob
 
no, that directory should already exist and have a bunch of Qt stuff in there. what happens if you use terminal.app to cd to that directory?
 
pfile said:
you can't open a module directly in PI. it has to be opened from the "install modules" menu. if it doesn't show up there, it means some dependency for the module has not been satisfied. in this case it is probably the tensorflow libraries. try quitting PI and moving the tensorflow libraries to

/Applications/PixInsight/PixInsight.app/Contents/Frameworks/

assuming you are using the libtensorflow* files that shipped with StarNetV2 and not the ones that ship with PI.

rob
Click to expand...

Please note that by doing this the PixInsight.app application bundle will be invalidated, since the digital signature won't be valid.
 
worked fine on my M1 macbook running monterey, PI launches just fine, and the OS doesn't complain.
 
I moved all the tensor flow libraries. There were two of them libtensorflow_framework..2.dylib and libtenosorflow2.dylib . I checked my PixInsight and StarNet is not showing up. I assume I need to do something else so please advise.
 
Mike - Thank you, thank you, I finally got it work! It was painful with all the errors I kept getting but step by step I worked around them until I was able to get the module to install (including over riding the trusted file error message). In any case it is possible to get it to work on a MacBook Air M1.
 
yeah, the problem always seems to be that the OS wants to disallow loading of the tensorflow libraries and the starnet module because they are unsigned. and it seems that sometimes you don't get a dialog box stating that the load was prevented, so you don't know to go over to the system preferences and OK the libraries/module. you might end up having to do this 3 times.
 
pfile said:
ok, this is the same thing i saw on my M1 mac, and so i ended up putting my libtensorflow* files into /Applications/PixInsight/PixInsight.app/Contents/Frameworks/ and then everything was happy. in truth the dynamic loader should be looking in /Applications/PixInsight/bin for the tensorflow libraries but as you can see it did not look there. i don't know why that is, but it might be because the Starnet2-pxm.dylib is not asking the loader to look there.

rob
Click to expand...
Thanks so much! It finally worked. StarNet need to clearly state this in the readme file.
 
Yuxuanwang2009 said:
Thanks so much! It finally worked. StarNet need to clearly state this in the readme file.
Click to expand...

the way i figured this out was to edit the PixInsight.plist file to manually add the StarNet2 process to PixInsight. when PI tries to load the module at startup time, the OS throws an error saying it couldn't find the tensorflow libraries. when PI is "forced" to load the module this way, it sees this error message and presents it to the user in a dialog box. the error message gives the list of directories that the loader was searching in; the frameworks directory is mentioned in that list. so i tried putting the tensorflow libraries in there, and that worked.

this error message is apparently silently discarded when you try to install a module thru the PI "manage modules" mechanism.

but as juan points out above, moving the library into the PI application bundle invalidates the code signature on the PI bundle. it appears though that the OS doesn't care about that - maybe it only checks at installation time.
 
What a freaking mess. Ok I had to put all files in the bin folder AND the framework folder. Then while clicking 'search' for modules to install, also clicking the 'install anyway' in the system security. But I had to do this multiple times. It just wouldn't show up as an installable module. The pb file MUST be in the bin folder, not lib folder, or both, but after I finally installed, got a big error that it couldn't find the 'find checkpoint file' in the /bin folder.... so once I added that pb file to bin the all worked. FINALLY!


*** Error: Could not find checkpoint file in '/Applications/PixInsight/bin/'!
Reading swap files...
9291.291 MiB/s
<* failed *>

StarNet2: Processing view: M42_Orion_RGB_session_1_101degCW_1_0x_LZ3_NS_crop
Writing swap files...
789.101 MiB/s
 
the multiple times thru the security preferences is because OSX first gets angry about the starnet module itself, and next it gets angry about the libtensorflow library. so there's no real way around that save everything being signed with a developer certificate. juan has one and when he adopted starnet v1 he signed all the libraries and of course the module, so you never see this problem. nikita doesn't have an apple signing certificate, hence the hassle.

as for the weights, this is one reason why when Juan refactored starnet V1 he put in the wrench icon to let you point at the weights files. but nikita didn't use that codebase when he wrote V2. the location is hardcoded.
 
jedashford said:
I would suggest someone write up detailed install instructions. Or, even better, make it install like the others by adding a url and have the rest automatic. Someone buy him a cert, they're not alot... https://stackoverflow.com/questions/6257506/how-much-does-it-cost-to-get-a-code-signing-certificate-from-apple
Click to expand...

this is all in nikita's hands, i don't think he is interested - he's aware of the PI distribution method and chose not to distribute his module that way. i and others around in the community always helped him build the OSX binaries. he owns no apple machines and apparently does not want to.

there are detailed install instructions around on the web (probably on nikita's website). in fact, i wrote the original OSX README file for starnet v1, but i was not involved in starnet v2 so can't say much about it. i did discover the M1/ARM problem with the loader and the need to put the files in the frameworks directory, but again no one with the power to fix this seems to be interested in the problem.
 
You must log in or register to reply here.
Back
Top