Retina

Open-source astrophotography image processing, fully scriptable in Python.

Retina calibrates, registers, integrates and processes astronomical images. Its core is a Python library that runs headless; the interface is one of its clients, not its owner.

Everything is reachable from the built-in Python console. Opening and saving, creating a window or a preview, choosing the active view, applying a process, managing masks and history, arranging the docks, running a batch — all of it goes through the retina API. No feature is reserved for the interface.

Better: every gesture in the interface prints the equivalent Python, executable and copyable. You can see it in the console at the bottom of the screenshot above:

# ← GUI: app.open('C:\\Users\\jroma\\Desktop\\h_m51_b_s05_drz_sci.fits')
# ← GUI: app.set_viewport((4300, 6100), zoom=0.0404918)
# ← GUI: app.layout.open_process('AIDeconvolution')

You learn the API by clicking, and any sequence of clicks becomes a script.

• 141 processes — calibration, cosmetic correction, debayer, registration, integration, background extraction, deconvolution, denoising, stretching, colour calibration, star removal, photometry, astrometry, mosaics, HDR.
• Automated pre-processing (retina.pipeline) — point it at a folder of raw frames and it scans, groups, builds masters, calibrates, registers and integrates, with caching and resume.
• Inspection and comparison — frame selector, blink, linked views, before/after curtain, FITS header.
• Interactive tools — masks composited in the shader, handle-driven crop, clone stamp, click-to-measure PSF, manual pair registration, celestial readout.
• Script mode — a full-page Monaco editor with hover and signature help from the embedded IPython, plus a Script process that makes a script run undoable and replayable.
• Projects — a .retina file stores an entire session, undo history included.
• Bilingual — English by default, complete French, detected at startup.

Prerequisites: Python ≥ 3.11, a Rust toolchain, Node ≥ 20.

git clone https://github.com/jromang/retina && cd retina
python -m venv .venv && source .venv/bin/activate
pip install maturin
pip install -e '.[web,xisf,astro,project,dev]'

# Build the native core.
maturin develop --release

# Build the frontend into the Python package.
cd web && npm install && npm run build && cd ..

python -m retina.web          # server + native window

Optional GPU acceleration (Linux and Windows only — the CuPy wheel is tied to a CUDA branch, which is why it cannot go in the line above):

pip install -e '.[cuda]'      # CUDA 13; use '.[cuda12]' for a CUDA 12 driver

Without any interface, in pure script form:

python -m retina.run recipe.py        # no graphical dependency
python -m retina.pipeline /data/M31   # automated pre-processing, headless

Each release ships an MSI, published on the releases page.

The installer is not code-signed, so Windows SmartScreen will warn on first run: choose More info → Run anyway. Verify your download against the SHA256SUMS.txt published beside it — noting that MSI packaging is not bit-for-bit reproducible, since WiX embeds timestamps and generated GUIDs, so a checksum identifies an artifact rather than a version.

Every release is built by a public workflow (release-windows.yml) from the source of this repository, with public logs: nothing is built on a developer machine and uploaded by hand.

What the software does with your data: PRIVACY.md. In short, it will not transfer any information to other networked systems unless you ask it to.

pytest -q -m "not gpu"                     # Python: domain + server, headless
ruff check python tests scripts
cd web && npm test                         # vitest
cd web && npx playwright test              # end-to-end, through a real browser

The suite self-skips what is not installed, so install the extras above if you want it to mean something.

• ARCHITECTURE.md — the design: the four non-negotiables, the three layers, the object model, the pre-processing engine, the web shell, GPU dispatch, packaging, and the traps worth knowing before you touch any of it.
• Per-process reference documentation ships with the application (help panel), in English and French, under python/retina/resources/doc/.

Issues and pull requests are welcome. Read ARCHITECTURE.md first — in particular the console/GUI parity rule, which is the constraint most likely to make a patch need rework: any capability added to the interface must exist in the scriptable API first.

Source code, comments and commit messages are in English. French exists in the product catalogues only, and python scripts/check_english.py enforces the boundary.

GPL-3.0-or-later.

Retina bundles third-party components; app.credits() in the console, or Help → Licences in the interface, lists every one with its licence. One caveat worth stating up front: the optional GraXpert AI models are CC BY-NC-SA 4.0 — free to use, but commercial use is prohibited. That restriction comes from the models, not from Retina, which restricts nothing.

Version française du README