91 replies to this topic

[p7ayfu77]

Hi all,

 

Id like to introduce a new open source tool and framework I've been working on for the last few months.

I guess pictures speaks louder than words, but a bit of the background is here: https://github.com/p...ise2Noise.ipynb

 

This was prompted by my interest in deep learning and some inspiration from the following threads:

 

* https://www.cloudyni...trophotography/
* https://www.cloudyni...se-attenuation/

 

Without further ado:

AstroDenoisePy: Before

AstroDenoisePy: After

 

The platform is quite versatile, allowing you to progress through to building your own denoise models from your data, or just use the GUI and a standard model I have trained with my data to denoise your images. In my testing so far even my "standard" model is very effective on other data I have not trained my models on.

 

Much more info can be found at the github repo with examples, and some background to how I got here: https://github.com/p...7/astro-csbdeep

This effort has been helped in part by friends in the Astrobiscuit community on Discord, and as such I want to thank all that have contributed with testing so far.

 

I have made an MSI distributable available here for a short period here: https://1drv.ms/f/s!...QwBOjlnhtu7_GkQ

Please note this is a small endeavour and so please be kind with requests or feedback, I will not be able to respond always.
I do hope this is useful for people.

Thanks
 

 

 

[pyrasanth]

I'm really grateful for your efforts. I've not tested the software however your screen shot looks like its very good indeed.

 

Do images being denoised need any masks?- however some testing will reveal that need if required.

 

I'm inspired that you have shared your insight with us.

 

Thanks.

[scottdevine]

What a great contribution. I’ll be sure to download and give it a try.

Thank you!

[Mert]

Looks great and that's a lot of effort!

Thank you for sharing!

I downloaded and installed on my laptop but have some problem with the screen

not showing in it's total extent.

Is there some way to start the program to show correctly on a laptop, or

could the lefthand menu be scrolled?

 

CS,

Mert

 

Attached Thumbnails

• 

[santafe retiree]

Looks great and that's a lot of effort!

Thank you for sharing!

I downloaded and installed on my laptop but have some problem with the screen

not showing in it's total extent.

Is there some way to start the program to show correctly on a laptop, or

could the lefthand menu be scrolled?

 

CS,

Mert

Same -- not scaling to screen resolution and no scroll bar to access entire menu section
 

[p7ayfu77]

Looks great and that's a lot of effort!

Thank you for sharing!

I downloaded and installed on my laptop but have some problem with the screen

not showing in it's total extent.

Is there some way to start the program to show correctly on a laptop, or

could the lefthand menu be scrolled?

 

CS,

Mert

Hmm this looks like it might be a DPI issue...
Hard to investigate, but I'll take a look.
For the time being could you change ur screen DPI and see if it helps?

[Mert]

Will do and let you know,
however it is a bit strange that the aplication seem to have a different resolution then my screen.

Ok, just checked but nothing helps.

My screen resolution is 1366x768 and the aplication looks like having 800x600 or alike.

 

Hope you can solve it!!

Best regards,

Mert

Edited by Mert, 26 February 2022 - 05:45 PM.

[avarakin]

Looks great!

Any thoughts on integrating this into PixInsight?

Also do you think it would be possible to use AI for deconvolution/sharpening?

[KlausKlaus]

Be careful w/ software of unknown origin…

[p7ayfu77]

Will do and let you know,
however it is a bit strange that the aplication seem to have a different resolution then my screen.

Ok, just checked but nothing helps.

My screen resolution is 1366x768 and the aplication looks like having 800x600 or alike.

 

Hope you can solve it!!

Best regards,

Mert

I've uploaded a new version with some GUI "fixes", really just tried to make things a bit smaller to fit into 768. Hope it helps. 

Uploaded new build "0.3.2-alpha", see first post for link.
 

https://github.com/p...89936a0d0d16e88

Tidy up GUI, separate common areas by color
Add snapshots feature, comparison tool
Add support for native file open/save dialog for platform: win32
Add selector for GPU/CPU
Always save TIF as unit16
Add version to app title

[p7ayfu77]

Uploaded latest release with a few GUI enhancements and small bug fixes.

 

Release 0.3.3-alpha and future releases can be downloaded from the GitHub releases section of the repo here:

 

https://github.com/p...sbdeep/releases

 

 

 

[Borodog]

Wow I can’t wait to try this. Thanks for developing and sharing.

[Manitu]

Hi to all

 

I would like to know how I could build a denoise model and train it with data. How does this work and how can I store a model.

 

May be someone knows?

 

CS Tom

[Robowarrior834]

Hi p7ayfu77,

 

Thank you for the app, the only thing I noticed is that there seams to be a memory leak when using a GPU. After the Denoise the app does not free up the GPU Memory, so the next time it tries running it is bottle necked.

 

Sincerly,

Robowarrior

Edited by Robowarrior834, 24 May 2022 - 09:08 AM.

[starhopper-de]

Hello p7ayfu77,

 

I do not have Windows any more. I tried it with wine and a windows VM but with no success.

Is is possible to compile it for Linux? Or can I start it from a python2 or python3 shell?

Thomas

Edited by starhopper-de, 12 June 2022 - 02:47 PM.

[Borodog]

Bump.

 

Any tips on the settings? I can't seem to find any documentation on github.

[benula]

I was able to get this running on linux after fixing a few things and the results absolutely blew my mind.

 

I submitted a PR to get my changes merged back, but if you want to try it in the meantime you can do:

git clone git@github.com:bgilsrud/astro-csbdeep.git

git checkout linux_fixes

 

Then read the README.md file as I added some documentation on how to run it on linux.

 

One thing that I think I learned, was that if you're denoising an image that has already been stretched then deselect the "Normalize" box.

 

Thanks p7ayfu77!!

 

-Ben

[Stanty8]

OP, your program is amazing! Far better results than Topaz for me. Any way we can buy you a coffee, metaphorically speaking? 

 

For those who want to try it out, there's a windows MSI file here: https://github.com/p...sbdeep/releases

[dciobota]

Absolutely fantastic program!  Now I finally have a Helix image I'm reasonably happy with.  Here is a sort of comparison.  The version on the left is my previous final image, and the one on the right is my new one. 

 

I processed it similarly to my previous image except I applied denoise before the post processing (the base image for both finished images was already preprocessed and stretched in APP) and was subsequently able to push the fainter regions a bit more.

 

Love this program, and I'll buy the developer a beverage of choice.  Failing that, I'd gladly put some coin in the donation box if one gets put up.  Outstanding work.

 

[Tapio]

Could there be a compare button to switch between original and denoised image, or a preview windows to compare ?

[dciobota]

Could there be a compare button to switch between original and denoised image, or a preview windows to compare ?

That's a tremendous idea.  The way I do it now is check/uncheck the denoise checkmark, but it's pretty slow.

[Devonshire]

I'm trying to understand where this is re: linux support for us non-python-developers.  Yes, I was in IT back in the day, but in Enterprise packageware.  Different world.  Possibly a different galaxy... :-)

 

Looking at github and trying to interpret, I *think* bgilsrud's linux changes may not have been merged back with p7ayfu77's changes.  Not sure. Also not sure about whether the GPU memory leak has been repaired.  And then there's kalgecin's astro-cssdeep, which talks about only denoising the luminance layer.

 

I'd love to take this for a spin on Mint, but it's really not clear which of these branches is best, and how to proceed from there.

 

Got advice?

[Manitu]

You can compare the original and the denoised one by clickung on the eye symbol on the lower right corner. It switches back and forth very fast.

[Devonshire]

I was able to get this running on linux after fixing a few things and the results absolutely blew my mind.

 

I submitted a PR to get my changes merged back, but if you want to try it in the meantime you can do:

git clone git@github.com:bgilsrud/astro-csbdeep.git

git checkout linux_fixes

 

Then read the README.md file as I added some documentation on how to run it on linux.

 

One thing that I think I learned, was that if you're denoising an image that has already been stretched then deselect the "Normalize" box.

 

Thanks p7ayfu77!!

 

-Ben

Ben,

 

I'm having a bit of trouble with the above - stuck partway.  

 

The above git clone command assumes that you have gone through the steps to set up an account on git.  I don't have one, so it fails.  

 

 

I did look look up the (publickey) error, and the suggested remedies all seemed to be in the context of maybe-this-or-maybe-that, for developers who wished to check in.  I just want to download this and try it - I doubt I'd ever have a reason to check in.

 

I WAS able to find an https: solution that got me through the clone without the account setup, but I'm stuck at the next step, the "checkout linux-fixes" line.   No idea how to deal with that.  

 

 

Can you help?

 

Thanks!

Edited by Devonshire, 12 November 2022 - 09:36 AM.

[Tapio]

You can compare the original and the denoised one by clickung on the eye symbol on the lower right corner. It switches back and forth very fast.

Thanks for this.