finding
active
finding:two-dinov3-blocks-track-scene-luminance-and-cast-shadow-volume-across-twelve-blender-models-with-block-norm-following-ground-truth-luminance-at-every-elevationTwo DINOv3 blocks track scene luminance and cast-shadow volume across twelve Blender models, with block norm following ground-truth luminance at every elevation.
Lighting/shadow manifold.
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.
Hypotheses (1)
hypothesis
- Motivates lighting search.
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.
- On DINOv3, classification and segmentation tolerate R²≈0.8, while depth needs R²≈0.9 to retain 95% accuracy.finding0.785Task-dependent distortion floors.
- Fig. 10.
- App. E.
- Concept dimensionality.
- Suggests architectural variations influence persona localization pattern
- Establishes generalizability of the core difficulty-boundary finding across model families.
- Supports claim that cross-modal alignment predicts downstream language task performance