Extracts dense 3D Signed Distance Fields from images in under 3 seconds using feed-forward geometry transformer latents.
March 30, 2026
Original Paper
Fus3D: Decoding Consolidated 3D Geometry from Feed-forward Geometry Transformer Latents
arXiv · 2603.25827
The Takeaway
Traditional 3D reconstruction requires costly per-scene optimization or post-hoc fusion; this method decodes geometry directly from pre-trained transformer embeddings, enabling near real-time 3D perception from unstructured images.
From the 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 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 disc