Benchmark of perspective video clips cropped from 360° footage for evaluating camera pose estimation in SfM methods.
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.
ORBIT++ is a research benchmark for Structure-from-Motion (SfM) and camera pose estimation, from Google DeepMind and collaborators. It builds challenging test clips from online panoramic 360° video: camera motion is tracked across the full panorama, then selected regions are cropped and reprojected into perspective-view clips with ground-truth trajectories. The panoramic context provides more reliable tracking even when parts of the view are blurred, moving or dynamic.
The project page reports that COLMAP and recent optimization-based and feed-forward methods (MegaSaM and VGGT-Long are shown) struggle on the benchmark. Results are broken down by six challenge categories: high speed, low texture, low light, crowds, egocentric parallel-moving objects, and fluids. Lack of texture is reported as the hardest for MegaSaM and COLMAP, and high speed for the other methods. Metrics include ATE and RPE-R.
ORBIT++ is an extended version of an earlier, smaller ORBIT benchmark published at CVPR 2026. The page gives a BibTeX entry (arXiv 2608.22039) but no download links, code or license in the provided material, so availability of the dataset itself is unconfirmed here.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Benchmark of hard perspective videos cropped from real 360° footage to test camera-motion estimation and 3D reconstruction. Existing systems like COLMAP and MegaSaM often fail on it.
Closest neighbours by meaning, not just by category.
Research
Research framework for globally consistent 3D reconstruction from long RGB videos using chunked priors and test-time adaptation.
Model
3D-aware 360° video diffusion model that generates panoramic videos along user-defined camera paths from sparse panoramas.
Model · Google
Predicts depth, surface normals and sky masks from a single 360° panorama using a cubemap-adapted 3D foundation model.
Research · Google DeepMind
Google DeepMind research that turns a text-to-video diffusion model into a single feed-forward, multi-task vision model.
Model
Feed-forward 3D Gaussian Splatting model that reconstructs scenes from arbitrary input views, with or without camera poses.
Model
Research method that recovers 6-DoF object and camera pose from monocular video via proxy-video translation and PnP.