Post-processing

Two small, deliberately pluggable stages in post_processing/. Each is driven by a Pydantic config, dispatched through a registry keyed by a method enum, applied marker-by-marker with a progress bar, and returns a new Trajectory3d carrying the original reprojection-error arrays through untouched.

Interpolation (interpolation/) fills gaps in the triangulated array. InterpolationConfig offers exactly one method today, linear: the marker's (frames, dims) slice goes through pandas.DataFrame.interpolate(method='linear', axis=0). Afterward fill_in_nans() replaces any surviving non-finite values (leading/trailing NaNs linear interpolation cannot reach) with the mean of all finite values in that marker's slice (a single np.nanmean call over the whole (frames, dims) array, not per column).

Filtering (filters/) smooths the interpolated signal. FilterConfig defaults define the pipeline's standard treatment: a butter_low_pass zero-phase Butterworth (SciPy butter + filtfilt) with cutoff 6 Hz, sampling rate 30 Hz, order 4, applied independently to each marker coordinate column.