Feed-forward 3D Gaussian Splatting model that reconstructs scenes from arbitrary input views, with or without camera poses.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
InfiniSplat v2 is a research model from Zhejiang University's 3D vision group (zju3dv) that reconstructs 3D Gaussian Splatting scenes in a single feed-forward pass from an arbitrary number of input views. It accepts images with or without known camera poses. The accompanying paper is titled "Analyzing and Improving Feed-Forward 3DGS from Any Views".
The method uses a dual-branch architecture that freezes geometry and learns Gaussian attributes, with opacity used to suppress artifacts from inconsistent points. Training uses scale-aligned ground-truth cameras and large-angle target views, which the authors say improves rendering under large viewpoint changes. The project page shows object and scene results, comparisons of reconstructed Gaussians against point clouds, and large-scale reconstruction on KITTI with more than 1,000 frames.
A YouTube creator describes code and models as released (about 6.6 GB), but the provided pages do not include a repository, Hugging Face link, or license, so those details are unconfirmed. It is aimed at researchers and developers working on 3D reconstruction, not at end users.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
3D reconstruction model that turns multi-view photos into explorable 3D scenes, including large driving-footage environments, with cleaner results than other methods. About 6.6 GB, code and models released.
Closest neighbours by meaning, not just by category.
Model
Feed-forward 3D scene reconstruction model that outputs triangle primitives directly usable as meshes for simulation.
Model · Nvidia
NVIDIA research model that turns a single image into a queryable 3D Gaussian scene with semantics and object instances.
Research
Research method that reconstructs complete 4D (3D plus time) scenes from a single monocular video using Gaussian Splatting.
Research
Post-training method that refines feed-forward 3D reconstruction models (Pi3, DA3, VGGT) on unlabeled video.
Model · NVIDIA
NVIDIA's 117M-parameter looped transformer for multi-view 3D reconstruction of camera poses and dense geometry.
Model
Research framework that turns posed images with instance masks and 3D boxes into per-object meshes for cluttered scenes.
