Differentiable Self-Calibration for FPP

FPP systems drift: temperature, vibration, and handling silently invalidate the calibration that all downstream 3D accuracy depends on. It is another hidden failure mode of deployed perception. This two-paper series makes single-shot FPP self-calibrating.

The first paper formulates differentiable self-calibration with uncertainty propagation: calibration parameters are recovered through the same differentiable pipeline that produces depth, and calibration uncertainty is propagated to per-pixel depth uncertainty, so the system knows not only what it measures, but how much to trust the measurement.

The second paper closes an identifiability gap: not every fringe pattern makes self-calibration well-posed. We use differentiable pattern design to optimize the projected patterns themselves so the calibration parameters are identifiable by construction.