Saved in:
Bibliographic Details
Main Authors: Fink, Laura, Franke, Linus, Kopanas, George, Stamminger, Marc, Hedman, Peter
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.25827
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910076953952256
author Fink, Laura
Franke, Linus
Kopanas, George
Stamminger, Marc
Hedman, Peter
author_facet Fink, Laura
Franke, Linus
Kopanas, George
Stamminger, Marc
Hedman, Peter
contents We propose a feed-forward method for dense Signed Distance Field (SDF) regression from unstructured image collections in less than three seconds, without camera calibration or post-hoc fusion. Our key insight is that the intermediate feature space of pretrained multi-view feed-forward geometry transformers already encodes a powerful joint world representation; yet, existing pipelines discard it, routing features through per-view prediction heads before assembling 3D geometry post-hoc, which discards valuable completeness information and accumulates inaccuracies. We instead perform 3D extraction directly from geometry transformer features via learned volumetric extraction: voxelized canonical embeddings that progressively absorb multi-view geometry information through interleaved cross- and self-attention into a structured volumetric latent grid. A simple convolutional decoder then maps this grid to a dense SDF. We additionally propose a scalable, validity-aware supervision scheme directly using SDFs derived from depth maps or 3D assets, tackling practical issues like non-watertight meshes. Our approach yields complete and well-defined distance values across sparse- and dense-view settings and demonstrates geometrically plausible completions. Code and further material can be found at https://lorafib.github.io/fus3d.
format Preprint
id arxiv_https___arxiv_org_abs_2603_25827
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Fus3D: Decoding Consolidated 3D Geometry from Feed-forward Geometry Transformer Latents
Fink, Laura
Franke, Linus
Kopanas, George
Stamminger, Marc
Hedman, Peter
Computer Vision and Pattern Recognition
We propose a feed-forward method for dense Signed Distance Field (SDF) regression from unstructured image collections in less than three seconds, without camera calibration or post-hoc fusion. Our key insight is that the intermediate feature space of pretrained multi-view feed-forward geometry transformers already encodes a powerful joint world representation; yet, existing pipelines discard it, routing features through per-view prediction heads before assembling 3D geometry post-hoc, which discards valuable completeness information and accumulates inaccuracies. We instead perform 3D extraction directly from geometry transformer features via learned volumetric extraction: voxelized canonical embeddings that progressively absorb multi-view geometry information through interleaved cross- and self-attention into a structured volumetric latent grid. A simple convolutional decoder then maps this grid to a dense SDF. We additionally propose a scalable, validity-aware supervision scheme directly using SDFs derived from depth maps or 3D assets, tackling practical issues like non-watertight meshes. Our approach yields complete and well-defined distance values across sparse- and dense-view settings and demonstrates geometrically plausible completions. Code and further material can be found at https://lorafib.github.io/fus3d.
title Fus3D: Decoding Consolidated 3D Geometry from Feed-forward Geometry Transformer Latents
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.25827