Configuration¶
The pipeline is configured via tile_compile.yaml.
Key Sections¶
input— source directory, frame limitsregistration— alignment engine, anchor strategy, astrometrymethod/aqmh— AQMH quality maps, reconstruction, storage, Cherry-Picktile— classic tile geometrystacking— weighting, rejection, output formatbackground_extraction— BGE parametersphotometric_color_calibration— PCC catalog, referenceruntime_limits— memory, time, worker limits, acceleration backend
GPU backend¶
runtime_limits:
acceleration_backend: auto # auto | opencv_cuda | opencv_opencl | cpu
parallel_workers: 8
memory_budget: 2048
auto prefers CUDA, then OpenCL, then CPU. Backend support is phase-specific:
PREWARP, AQMH maps, classic tile reconstruction, synthetic reconstruction, and
STACKING support CUDA/OpenCL; streaming AQMH reconstruction currently supports
CUDA; REGISTRATION remains CPU-only. AQMH Cherry-Pick deliberately uses CPU.
The effective choice is written to artifacts/acceleration_context.json and
shown in live progress logs.
Schema¶
Validate with:
Get schema:
Examples¶
See tile_compile_cpp/examples/ for scenario-specific configs:
m104.example.yaml— Alt/Az, strong rotation, poor seeingfull_mode.example.yaml— high quality equatorialemergency_mode.example.yaml— minimal runtimesmart_telescope_dwarf_seestar.example.yaml— smart telescopecanon_equatorial_balanced.example.yaml— balanced DSLR
Parameter Studio (GUI3)¶
The web frontend provides a guided parameter editor with:
- Scenario presets (altaz, rotation, bright stars, few frames, gradient)
- Situation Assistant — automated parameter suggestions
- Real-time validation against schema