Raviteja Garlapati

Gauss Lens: Neural perception, auditable geometry Gauss Lens extracts auditable 3D face parameters from a single photo, with no neural network in the 3D solve. Today’s neural face trackers return bare numbers, with no way to see where they came from and no way to refuse a bad photo. Gauss Lens splits the work in three. Meta’s SAM 3.1 turns the photo into masks and 27 landmarks. A plain NumPy Gauss-Newton solver fits the 3D face and runs a five-score quality gate, so every refusal names the score that failed. Meta’s Muse Spark writes a short verdict from the solver’s report only, and every number it quotes must match the solver’s. A test fails if any solver module imports an ML library.