AI Paper Atlas

Research map / 3D vision / Depth and stereo

On Geometric Understanding and Learned Priors in Feed-forward 3D Reconstruction Models

Jelena Bratulić, Sudhanshu Mittal, Thomas Brox, Christian Rupprecht

ECCV 2026 · 3D vision › Depth and stereo · approach: Cost-volume multi-view stereo · 13 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
zero out high correspondence-matching attention heads in middle layers (ablation)VGGT and DA3 · ShapeNet, 50mm focal lengthSampson distance errorHurt
zero out random early or late layers' attention heads (ablation)VGGT, DA3, DUSt3R · ShapeNet, 50mm focal lengthSampson distance errorNo clear effect
more local knockout (patch level instead of whole head) (ablation)VGGT, DA3, DUSt3R · ShapeNetSampson distance error degradationMixed
MLP probe vs linear probe for fundamental matrix recovery (ablation)VGGT, DA3, DUSt3R · fundamental matrix probingSampson errorImproved
mask part of one image with white patchesVGGT · ShapeNet pairs, small viewpointSampson errorNo clear effect
change lighting on only one viewVGGT · ShapeNet lighting variationsfailure rateHurt
add asymmetric lighting shadows to symmetric repeated scenesVGGT · ShapeNet repetition scenesgeometric accuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy