Documentation
This section collects technical notes and usage guides for StarNet and DeepSNR: PixInsight and command-line references, input behavior, data-quality guidance, and workflow examples.
DeepSNR PixInsight module reference
DeepSNR 1.3.0 remembers interface settings between sessions, alongside minor bug fixes and improvements. This reference covers settings, input preparation, and linear-data processing.
Remembered interface settings
The interface remembers its last-used options across PixInsight sessions. Use Reset to restore factory defaults. Imported process instances keep their own parameter values; saved interface settings do not change script defaults.
Input and processing guidance
Start with an integrated image before heavy processing. DeepSNR works best with uncorrelated noise; for Bayer-camera data, follow the input guidelines, including CFA drizzle integration where appropriate.
- Model v2 supports RGB and grayscale images and is the recommended starting point. Model v1 supports RGB only.
- Strength blends the denoised result with the original image.
- Linear data applies the current STF stretch, or an automatic stretch when no STF stretch is set, then returns the result to linear form.
Compare bright regions with the original when processing linear data; stretching and denoising can change their fidelity.
Troubleshooting
When reporting a processing problem, include the full PixInsight Process Console log. It identifies the module version and build, backend initialization or fallback, and total processing time. A GPU warning followed by successful CPU initialization means processing can continue on the CPU.
Related pages
- DeepSNR PixInsight module repoRepository URL and current TensorFlow/runtime deprecation note
- PixInsight installationIllustrated repository setup flow
- DeepSNR input guidanceCamera, channel, and noise guidance
- DeepSNR command-line toolCommand syntax, inputs, outputs, and model options