Research map / 3D vision / Depth and stereo
Latent Fusion: Decoding Consolidated 3D Geometry from Feed-forward Geometry Transformer Latents
Laura Fink, Linus Franke, George Kopanas, Marc Stamminger, Peter Hedman
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Regress SDF directly from transformer latents instead of fusing per-view predictions (TSDF) | VGGT_ft · Objaverse 8 views | F-score at 0.5 eps: 0.43 -> 0.51 | Improved |
| Regress SDF directly instead of TSDF fusion | VGGT_ft · Objaverse 8 views | SDF MAE: 0.023 -> 0.004 | Improved |
| Regress SDF directly instead of TSDF fusion | VGGT_ft · Objaverse 8 views | EMD: 0.014 -> 0.019 | Hurt |
| Regress SDF directly instead of TSDF fusion | VGGT_ft · Objaverse 8 views | Chamfer distance: 0.022 -> 0.021 | No clear effect |
| Replace Poisson fusion with latent SDF extraction | VGGT_ft · Objaverse 8 views | F-score at eps: 0.70 -> 0.83 | Improved |
| Replace per-view fusion with joint latent extraction at 24 views (trained on up to 8) | Fus3D · Objaverse varying views | F-score / Chamfer | Improved |
| Use Fus3D over UFORecon with VGGT poses (unfavorable views) | UFORecon · DTU unfavorable | Chamfer distance: 4.942 -> 3.525 | Improved |
| Use Fus3D over VolRecon with VGGT poses (favorable views) | VolRecon · DTU favorable | Chamfer distance: 3.678 -> 2.432 | Improved |
| Compare Fus3D with UFORecon given COLMAP poses (unfavorable) | UFORecon · DTU unfavorable | Chamfer distance: 3.275 vs 3.525 | Mixed |
| Use learned canonical embedding instead of projected voxel positions (variable query volume) (ablation) | Fus3D+ · Objaverse variable query | Chamfer distance: 0.021 -> 0.018 | Improved |
| Use learned canonical embedding (default query volume) (ablation) | Fus3D+ · Objaverse default query | F-score at 0.5 eps: 0.466 -> 0.467 | No clear effect |
| Add near-surface SDF supervision term (ablation) | Fus3D · Objaverse | isosurface accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UFORecon — UFORecon: Generalizable Sparse-View Surface Reconstruction from Arbitrary and Unfavorable Sets (CVPR 2024) · better on 2 of 2 shared results
- VolRecon — VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction (CVPR 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning Scene-Level Signed Directional Distance Function With Ellipsoidal Priors and Neural Residuals (TPAMI 2026 · similarity 0.88)
- Structured 3D Latents for Scalable and Versatile 3D Generation (CVPR 2025 · similarity 0.87)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.87)
- WildFusion: Learning 3D-Aware Latent Diffusion Models in View Space (ICLR 2024 · similarity 0.87)
- Tri-Perspective View Decomposition for Geometry-Aware Depth Completion (CVPR 2024 · similarity 0.86)
- Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis (NeurIPS 2026 · similarity 0.86)
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.
