AI RESEARCH
MUSE: Resolving Manifold Misalignment in Visual Tokenization via Topological Orthogonality
arXiv CS.CV
•
ArXi:2605.05646v1 Announce Type: new Unified visual tokenization faces a fundamental trade-off between high-fidelity pixel reconstruction (spatial equivariance) and semantic abstraction (conceptual invariance). We attribute this conflict to Manifold Misalignment: naive joint optimization induces opposing gradients, creating a zero-sum game between reconstruction and perception. To address this, we propose MUSE, a framework based on Topological Orthogonality.