Fus3D: Decoding Consolidated 3D Geometry from Feed-Forward Geometry Transformer Latents
Fink L, Franke L, Kopanas G, Stamminger M, Hedman P (2026)
Publication Type: Conference contribution
Publication year: 2026
Journal
Original Authors: Laura Fink, Linus Franke, George Kopanas, Marc Stamminger, Peter Hedman
Pages Range: 578-597
Event location: Malmö
ISBN: 9783032371669
DOI: 10.1007/978-3-032-37167-6_33
Abstract
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 (FFGT) 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 FFGT 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.
Authors with CRIS profile
Laura Fink
Chair of Visual Computing (LGDV)
Linus Franke
Chair of Visual Computing (LGDV)
Marc Stamminger
Chair of Visual Computing (LGDV)
How to cite
APA:
Fink, L., Franke, L., Kopanas, G., Stamminger, M., & Hedman, P. (2026). Fus3D: Decoding Consolidated 3D Geometry from Feed-Forward Geometry Transformer Latents. In Proceedings of the European Conference on Computer Vision (ECCV 2026) (pp. 578-597). Malmö, SE.
MLA:
Fink, Laura, et al. "Fus3D: Decoding Consolidated 3D Geometry from Feed-Forward Geometry Transformer Latents." Proceedings of the European Conference on Computer Vision (ECCV 2026), Malmö 2026. 578-597.
BibTeX: Download