finding
active
finding:a-dinov3-grassmannian-block-recovers-a-cabbage-cauliflower-manifold-with-smooth-feature-visualization-transitionsA DINOv3 Grassmannian block recovers a cabbage/cauliflower manifold with smooth feature-visualization transitions.
Fig. 10.
Source paper
extracted_from(2026) · Thomas Fel · Matthew Kowal · Mozes Jacobs · Dron Hazra +21
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Motivation for lifting atoms to blocks.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Interpretation of stable rank.
- MDL result at 20% distortion.
- Curve manifold recovery.
- Lighting/shadow manifold.
- Cross-modality result from the full paper demonstrating that representation-behavior geometry alignment is not limited to language models.
- On DINOv3, classification and segmentation tolerate R²≈0.8, while depth needs R²≈0.9 to retain 95% accuracy.finding0.729Task-dependent distortion floors.
- Block and Grassmannian featurizers keep utilization ≈0.6–0.66 at k=16, while Group Lasso falls to ≈0.17.finding0.726Per-feature occupancy.
- Central thesis.