GraXpert models — mirror for Retina
This repository is a redistribution mirror of the neural-network models produced by the GraXpert project, so that Retina can fetch them without requiring a full GraXpert installation.
These models are not ours. They are the work of the GraXpert Development Team and its
contributors, and they are licensed under Creative Commons Attribution-NonCommercial-ShareAlike
4.0 International (CC BY-NC-SA 4.0) — distinct from the GPL-3 licence of the GraXpert
application code. The files here are unmodified: they are the exact model.onnx extracted
from GraXpert's official model archives, and their SHA-256 sums are listed below so anyone can
verify that.
⚠️ NonCommercial
CC BY-NC-SA 4.0 forbids commercial use. You may download and use these models for non-commercial purposes only. If you sell prints, work on commission, or otherwise process images commercially, you are not licensed to use these models — this restriction comes from GraXpert, not from Retina, which itself places no such limit. Redistribution is permitted (and this mirror relies on it) provided attribution, the same licence, and no additional restrictions.
Attribution
- Authors: GraXpert Development Team and contributors.
- Source: https://github.com/Steffenhir/GraXpert
- Full licences and contributor lists (verbatim, in
licenses/):- Background extraction —
BGE-Model-LICENSE.html - Denoising —
Denoise-Model-LICENSE.html - Deconvolution —
Deconvolution-Model-LICENSE.html
- Background extraction —
- Changes: none. Files are byte-for-byte the models GraXpert distributes.
Models
| File | Task | Version | Size | SHA-256 |
|---|---|---|---|---|
graxpert-denoise-3.0.2.onnx |
Denoising | 3.0.2 | 478 Mo | e61c880e7fb95591… |
graxpert-background-1.0.1.onnx |
Background extraction | 1.0.1 | 218 Mo | 4ed3ef68c9f44e82… |
graxpert-deconv-object-1.0.1.onnx |
Deconvolution — extended object | 1.0.1 | 279 Mo | 36b88b422cf98521… |
graxpert-deconv-stellar-1.0.0.onnx |
Deconvolution — stellar | 1.0.0 | 279 Mo | c9a8a2a4fa40eaca… |
How Retina uses them
Retina prefers a local GraXpert installation if it finds one (its models are discovered and used in place). This mirror is the fallback for users who do not have GraXpert installed. In both cases the model actually used — its name, version and SHA-256 — is written into the processing history and the FITS keywords, so every result records exactly which file produced it.