title | emoji | colorFrom | colorTo | sdk | sdk_version | app_file | pinned | license |
---|---|---|---|---|---|---|---|---|
AstroSleuth |
๐ |
pink |
yellow |
streamlit |
1.17.0 |
app.py |
false |
gpl-2.0 |
The (only?) free, no-strings-attached, open source astrophotgraphy upscaler toolset. I wanted a solution that can run on almost any hardware with epic results, and I hope this repo serves you well.
If you ever want to support the project, please give the repo a star so that it's easier for others to discover it.
Quick start on colab, but checkout running for more details.
- Tips - for image samples and potential workflow improvements
- Training - for more information on each model
- AstroNEXT and AstroFAST models to replace V2 and V1 respectively. Read training for more details.
- Minor UI changes for reading model descriptions.
- Scaled the preview image for faster loading.
- Ability to 'coerce' NEXT into stars vs details. See conditioning.
- Most models convert large stars into 'star clusters'
- Diffraction spikes may turn into diffraction stars
- Fine nebula details may "smush" together (varies from image-to-image)
I started this project a regrettably long time ago. I tried different ways to share my work, got some hate, some love, and setted for what you see now.
I present an acculmination of multiple ideas, improvements, and lessons; trained on 15 thousand images of various astrophotography targets.
It is behind my works on reddit, my youtube attempt and my cloudy nights post, and I hope it leads the way for any other future attempts; for anyone.
A lot has happend throughout this project. I guess cause it was my first machine learning application and it's been a nice distraction every now and again.
Ask for any improvements and I will likely implement them. Any feedback is appreciated, such as creating a Photoshop/Pixinsight plugin? Just open a git issue here and I'll see to it.
This tool is presented as is, free as long as it stays free. "Scientific accuracy" was never the goal of this project - it was made to help others, and to make astronomy a little easier.