nekitmm / starnet Goto Github PK

how can i create sparate the photo to one starless photo and one starmask photo

Hi, just passing by on what seems to be a very cool projet. I saw a picture on instagram of someone using your code to remove stars and the effect was amazing.

While trying to test if myself, I run into a save error:

Which is linked to that line: https://github.com/nekitmm/starnet/blob/master/transform.py#L122

Probably removing the "./" would be a good idea as i was trying to run your script with an absolute path and not a relative path and it couldn't save due to a weird path :)

PS: also didn't managed to have the plugin installed with pixinsight 1.8, including with recursive search - even though the files were there. Maybe something changed with the versions ? It's not related to your warning on AVX since i could run the script successfully.

Anyhow, great job and interesting project !

Pixinsight
problem in batch treatement with image container and linear data. The linear data checked is not transmited at image container so black artifacts...
Thank you i love starnet2

Greetings,

Sorry I missed the train mention in the README.md file. My mistake.

Regards

Stephen

Hi @nekitmm

I just wanted to report that I've found a nice/simple plugin to integrate StarNet with GIMP (have the starless automatically imported as layer). I think it's totally worth the mention in your website.

You may find it here:
https://github.com/pigeond/gimp-plugin-starnet

PS: I didn't know about this repo... is this the source for the binaries in the website (I don't see releases here, hence my question)? If it's all Python, then the plugin could actually be importing the code - one thing that could definitely improved is the progress completion feedback to GIMP....

Hi! I'm not sure how to reach you except by creating an issue. If this is not the correct way, I apologize.

I enjoy starnet++ and wanted to contribute,
I've created a repository that enables users to use a containerized version of the Starnet CLI to process (or batch process) their images using Starnet++.
I would love for you to try it out & let me know what you think:
GH Repo: https://github.com/thijsvanloef/starnet-docker
Docker Hub: https://hub.docker.com/repository/docker/thijsvanloef/starnet-docker

Hi,
toimage from scipy.misc has been removed since version 1.2.
One alternative could be to use PIL instead :)

Hi Nikita,

actually working to train with new data using my instrument that generate stars with spikes. Slowly it seems going well. I was wondering how do you proceed to generate both .pb freezed files (L and RGB) model in order to use with PI?

thanks!

Hello,
I didn't find any interface for MacOS to run it outside Pix

Any guidance please to run it on MacOS ??

Something about the way I'm running this is causing an error in the reading of the log files. The later epochs that I have run seem to create a "Perceptual_losses.txt" with too many spaces in it, causing it to see too many columns in the later lines (epoch 500+)

Not sure what might be causing it to give more spaces in the log... what version of numpy did you use?

Looks like a great package.
Unfortunately it does run on the latest version of tensor flow as various changes
have been implemented to the package.
I tried hard but could not get it to run.
Please comment here if anyone has been more successful.
Thanks, Des

Is there a build process for making our own release of a PI module to experiment with?

I've looked everywhere and I can't find sources for Starnet++ binaries published on SourceForge, only the Python files in this repo.

Is the code those binaries are built from not open-source? If it is, could it be added to this repo as well?

Getting error below

Anaconda3\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 64, in
from tensorflow.python._pywrap_tensorflow_internal import *
ImportError: DLL load failed while importing _pywrap_tensorflow_internal: The specified module could not be found.

Failed to load the native TensorFlow runtime.

See https://www.tensorflow.org/install/errors

for some common reasons and solutions. Include the entire stack trace
above this error message when asking for help.

Hello Nekitmm:

Thanks your work! But there was an error happened during my installation of StarNet on my Mac OS X 10.15:
Log:
Installing 1 module(s):
/Applications/PixInsight/bin/StarNet-pxm.dylib
<* failed *>
*** PixInsight API Error: Module: /Applications/PixInsight/bin/StarNet-pxm.dylib: Unsupported API version 162 (expected = 160).: Module load error

What this error means and how could I fix it?

Hello,

I am an astrophysicist from team working with University of Waterloo (UW), Canada. We find this project very interesting and of potential usage for our big ongoing full-sky survey. We would like to experiment with improving starnet performance a little (because unfortunately it is not always produce adoptable results) and probably tune it for FITS we will use. Could you share your expertise about data you train on? Specifically:

• Can you share the data used and if yes, on which terms?
• How big the data size for training you use?
• How do you remove stars from images in training data to train the whole net?

In any case we will surely acknowledge and referencing your work

hi nekitmm， i saved a milky way core tif in photoshop in 16 bit, but when i go to starnet++, it always say only support 16 bit files, any idea?

Hi,
Love and appreciate the work you put into this. It's an awesome tool!

Recently I've been trying to retrain it to be able to remove the alterations I get with a reflector telescope. I've been trying to get others interested enough to give me some sample images so that I can do a nice long training run...

Do you have any advice on the types of image that will improve the training? I first trained it for 70 epochs with 5 whole images that I prepared, but the output images weren't getting any better at removing the diffractions spikes, which I figured was because it was seeing them too infrequently on the whole images. After that, I tried using smaller images where I had just had clipped out the stars with spikes, but the output I'm getting now is very blocky...

I'd appreciate any advice on the type and number I'd need for a good training set! I'm just starting again now from fresh wit the 5 original images... on Epoch 50...

On running starnet I get the following error:

adrian@mononoke:~/starnet$ python starnet.py transform ../Flat/Keeps/M42-and-M43.png
Traceback (most recent call last):
File "/home/adrian/starnet/starnet.py", line 105, in
import transform
File "/home/adrian/starnet/transform.py", line 19, in
from scipy.misc import toimage
ImportError: cannot import name 'toimage' from 'scipy.misc' (/usr/lib/python3/dist-packages/scipy/misc/init.py)

My python installation is python 3.9.7 and scipy 1.6.0.

It appears that the toimage function in scipy was deprecated and then removed, the official recommendaton is to use a function from the pillow library instead (Image.fromarray).

I edited the python files as you suggested in replying to the other issue I raised and got rid of the toImage errors, unfortunately I now get a variety of tensorflow errors instead that I can't make go away by importing tensorflow.v1.compat and disabling v2 features.

Hello,
First of all, that's a very cool project, for me it works great, it helps to remove stars easily from astrophotos for easier nebulosity processing. I wish the weights were stored in this GitHub repository (maybe on master branch, maybe on separate branch) for easier updating. If you would update the weights, users would have to download and update them manually, but if they were stored in this repository, updating them locally would be as easy as typing git pull! 😄

You'd have to use https://git-lfs.github.com/ because there's a 100MB/file limit on regular repos

It's returning a 404. Thanks for making this project!