Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a novel approach to blind image restoration by modeling image degradations on a low-dimensional Riemannian manifold, allowing for a more accurate representation of the underlying degradation geometry. By formulating the restoration process as geodesic transport in the joint image-manifold space, the authors leverage a geodesic flow matching objective to learn intrinsic transport dynamics that account for the curvature of the degradation space. The results demonstrate that this method not only generalizes linear flow matching but also effectively handles mixed degradations, offering a robust theoretical foundation for improved generalization in restoration tasks.
Geodesic flow matching reveals that blind image restoration can achieve superior results by respecting the curvature of degradation spaces, challenging traditional linear assumptions.
Blind image restoration requires recovering clean images from observations corrupted by unknown and potentially mixed degradations. While recent deterministic flow-based methods model restoration as transport processes that map degraded images to clean ones, they typically rely on Euclidean interpolation, implicitly assuming linear degradation geometry. In this paper, we explicitly model degradations as points on a low-dimensional Riemannian manifold and formulate restoration as geodesic transport on the joint image-manifold space. Using a geodesic flow matching objective, we learn intrinsic transport dynamics that respect the curvature of degradation space. This framework generalizes linear flow matching, provides a principled treatment of mixed degradations as geodesic compositions, and yields a clean theoretical interpretation for generalization beyond observed degradations.