Focusing has been the biggest challenge for amateur astro-photographers.
This is a new and experimental approach to achieve a dead-correct focus with minimal steps using image-based deep learning
Currently
- it has numerous machine learning approaches.
- it has star-field simulations for arbitrary refractors and C8-like SCTs.
- it is using simulated focuser with mechanical errors
The goal is not relying on any base/foundational model.
- Simulate highly correct random Star field images
- Build multi-gpu train and hyper-parameter tuning environment
- Find working foundational approach, including agent (image - slm - fft)
- Connect with indilib
- Package Software out of scripts
- GUI for try out
- Release v1.0
- Distillation for tiny edge devices (RPi)
- Promote
Note, currently profilers are everywhere (for good reason)
NASA/ESA projects are used to simulate images without taking actual star photos (in the middle of cold nights)
Personal correct focus and out-of-focus images can be used
GPLv3+ (look here ASI/ZWO people)