Feed-Forward 3D Reconstruction and Visual Geometry
Recovering camera poses, point clouds, matches and motion from images in one network pass instead of slow optimisation.
Tier Grades
Tier A: Frontier AI labs
feasible
Meta and NAVER LABS lead feed-forward geometry models such as VGGT and DUSt3R.
Tier B: Indonesian AI labs
stretch
Training such a model is a stretch for an Indonesian lab; applying released ones is feasible.
Tier C: Institution units such as ITB
feasible
A 6-author team produced VGGT, a size a university unit can match for follow-up work.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Released models turn photos into 3D scenes quickly for site survey and inspection demos.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeTraining compute is not in the sources read.
- DataNot in the sources read; VGGT infers cameras, point maps, depth and tracks from one, a few or hundreds of views.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 6 authors on the VGGT paper (arXiv 2503.11651).
- Time horizonMonths to a year.
- Shelf-life signalsOne feed-forward model replaces structure-from-motion pipelines, Metric reconstruction works from phone video
- Pipeline roleTier A releases geometry models; Tier C evaluates them on local scenes; Tier D uses them for 3D site capture in proofs of concept.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Links
R&D TopicFeed-Forward 3D Reconstruction and Visual Geometry
Use Casesnone recorded
Who Runs Itnone recorded
Papers
- DUSt3R: Geometric 3D Vision Made EasyShuzhe Wang et al., arXiv preprint, 2023 founding
- VGGT: Visual Geometry Grounded TransformerJianyuan Wang et al., CVPR 2025 (per arXiv comment), 2025 state of the art
- Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward PassJianing Yang et al., CVPR 2025 (per arXiv comment), 2025 state of the art
- Grounding Image Matching in 3D with MASt3RVincent Leroy et al., arXiv preprint, 2024 state of the art
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D ReconstructionDavid Charatan et al., arXiv preprint, 2023 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2503.11651 https://arxiv.org/abs/2503.11651
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Feed-Forward 3D Reconstruction and Visual Geometry", https://aurion.semai.id/topics/feed-forward-3d-reconstruction/, records of October 2026.