Information-Regularized Constrained Inversion for Stable Avatar Editing from Sparse Supervision

TL;DR

This paper introduces a novel constrained inversion method for stable avatar editing from sparse supervision, utilizing a structured latent space and conditioning objectives to prevent identity leakage and temporal flicker.

Contribution

It proposes a conditioning-guided reconstruction framework that restricts edits to low-dimensional subspaces and optimizes constraints based on a local linearization of the rendering pipeline.

Findings

Improves stability of avatar editing with limited supervision.

Operates efficiently on small subspace matrices.

Predicts edit stability using an information matrix spectrum.

Abstract

Editing animatable human avatars typically relies on sparse supervision, often a few edited keyframes, yet naively fitting a reconstructed avatar to these edits frequently causes identity leakage and pose-dependent temporal flicker. We argue that these failures are best understood as an ill-conditioned inversion: the available edited constraints do not sufficiently determine the latent directions responsible for the intended edit. We propose a conditioning-guided edited reconstruction framework that performs editing as a constrained inversion in a structured avatar latent space, restricting updates to a low-dimensional, part-specific edit subspace to prevent unintended identity changes. Crucially, we design the editing constraints during inversion by optimizing a conditioning objective derived from a local linearization of the full decoding-and-rendering pipeline, yielding an edit-subspace information matrix whose spectrum predicts stability and drives frame reweighting / keyframe activation. The resulting method operates on small subspace matrices and can be implemented efficiently (e.g., via Hessian-vector products), and improves stability under limited edited supervision.