Event
FlatVPR: Geometric Correction of Foundation Model Feature Manifolds
1 reports · 1 sources
Rai Hisada and Kanji Tanaka propose FlatVPR, a plug-and-play geometric linear residual adapter: it applies a residual transformation to foundation model features such as DINOv2-ViT-S/14, and uses a Pullback Flatness Loss to reduce the curvature of the feature manifold, so that descriptors between adjacent anchors can be reconstructed by linear interpolation. The authors state that on the NCLT dataset, even when the anchor spacing reaches 100m and severe seasonal changes are present, applying this adapter still brings significant performance improvements. Map construction is incorporated into an EM framework, divided into a continuous M-step for manifold adaptation and an E-step for the anchor selection criterion.
Report timeline
- FlatVPR: A plug-and-play geometric linear residual adapter for geometric rectification of foundation model feature manifolds
FlatVPR proposes a plug-and-play geometric linear residual adapter that applies a residual transformation to features from foundation models such as DINOv2-ViT-S/14 and uses a Pullback Flatness Loss to reduce manifold curvature, so that descriptors between adjacent anchors can be reconstructed by linear interpolation. On the NCLT dataset, the adapter still delivers clear performance gains even when anchor spacing reaches 100 m and severe seasonal changes are present. Map construction is incorporated into an EM framework, split into a continuous M-step for manifold adaptation and an E-step for the anchor selection criterion.
arXiv cs.RO ·