Photometric color calibration node (SPCC / PCC)

[bscholer]

Why

SCNR-green (color_balance node) kills the dominant green cast on OSC stacks, but it can't fix an underlying R-heavy / B-light pedestal — the result is a neutralized-green but yellow-tinted sky. The proper fix is photometric color calibration: pick stars, look up their physical colors, derive a 3-channel gain transform, apply.

Siril 1.4 ships two relevant commands:

• pcc — older, uses NOMAD or APASS, decent for broadband OSC.
• spcc — Siril 1.4 successor, uses Gaia DR3 spectra + camera spectral response curves. Markedly more accurate.

Either applies a 3x3 color transform to every pixel (so background sky also goes neutral as a side effect, not just stars).

Catalog problem

Siril's spcc wants Gaia DR3 photometry locally. Full DR3 is ~hundreds of GB on disk; we are not shipping that in a Docker image. Options worth weighing:

1. On-demand Vizier / Gaia archive query (recommended starting point)

• Fetch only the stars in the FOV from https://vizier.cds.unistra.fr or https://gea.esac.esa.int/archive on each spcc call.
• Typical Dwarf 3 FOV: ~3° square, ~10k stars from Gaia DR3 — ~5 MB per query.
• Pros: zero install footprint, always fresh, works in the CPU Docker image.
• Cons: needs internet, subject to CDS rate limits (~10 requests/sec is fine), one extra failure mode.
• Cache responses keyed on (RA, Dec, radius) under /scratch/astrolab/cache/_catalog/ so reruns are free.

2. Lazy local subset

• First spcc call fetches the chunk for that sky region (HEALPix tile, ~50 MB per tile at NSIDE=64) and writes it to a Gaia mini-catalog on disk.
• Subsequent renders in the same patch of sky hit local. Grows organically up to maybe a few GB for a casual hobbyist.
• Pros: offline after first use per target.
• Cons: more state, needs an admin endpoint to clear it.

3. APASS DR10 fallback

• ~700 MB, B-V photometry, no spectra. Plain pcc works against it.
• Pros: small enough to ship in the GPU image, decent for OSC broadband.
• Cons: less accurate than SPCC, won't satisfy the long-term "do it right" goal.

4. NOMAD via Siril

• Siril 1.4 already knows how to fetch NOMAD on demand (small footprint).
• Pros: zero work from us.
• Cons: NOMAD is older and less accurate than Gaia/APASS.

Proposed shape

New color_calibrate node (or extend color_balance with a mode: scnr | spcc | pcc enum):

• Input: image (IMAGE_FITS, plate-solved; should run after register).
• Params:
  • mode: scnr (current behavior) | spcc | pcc
  • catalog: gaia_vizier | gaia_local | apass | nomad (only relevant for spcc/pcc modes)
  • narrowband_mode: bool, for emission-line workflows
• Implementation: shell out to Siril pcc / spcc with the right args; expect plate-solved input.
• Cache key: include catalog choice so flipping catalog rebuilds.

Recommendation for v1

Start with option 1 (Vizier on-demand) + APASS fallback (option 3). Vizier is the right "hosted Gaia" answer with no install pain. APASS as a no-internet fallback covers offline use. Add option 2 later if rate limits or latency bite.

Out of scope for this issue

• Siril SPCC's spectral response curves for the Dwarf 3 sensor — Siril ships a few presets but a Dwarf-specific curve is its own follow-up.
• Switching the template default; this node lands as opt-in until the catalog story is solid.

Related

• Filed after the color_balance (SCNR-green) node landed in Add color_balance (SCNR green) node to kill OSC green cast #113 cleared the green cast but left a yellow pedestal on Veil rendering, which is exactly the gap SPCC fills.