Predicts depth, surface normals and sky masks from a single 360° panorama using a cubemap-adapted 3D foundation model.
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.
PaGeR (Panoramic Geometry Reconstruction) is a research model, presented as a NeurIPS 2026 paper, that estimates panoramic geometry from a single equirectangular 360° image. It starts from a pre-trained perspective 3D foundation model (Depth Anything 3, ViT-Giant) and adapts it to panoramas with minimal architectural changes. The panorama is projected to a fixed 6 × 504 × 504 cubemap, and the six faces are treated as a multi-view input set.
In a single forward pass, four heads output scale-invariant depth, surface normals, a sky mask and a coarse-metric depth that rescales depth into metres. The model accepts both perspective and omnidirectional images. According to the project page, it outperforms the compared methods on scale-invariant depth, on most metric depth measures, and on surface normals on the benchmarks reported.
The authors also release two datasets: PanoInfinigen, a synthetic set of about 70,000 4K panoramas from 20,000 scenes plus roughly 7,000 urban panoramas, with ground-truth metric depth and normals; and ZüriPano, a real-world outdoor benchmark of 100 LiDAR-scanned 8K panoramas from Zürich. A YouTube creator reports that the model (about 6 GB) and datasets have been released. The exact license, code repository and weights location were not confirmed in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Panoramic Geometry Reconstruction model that predicts depth and surface normals from a single 360° panorama by adapting a 3D foundation model to cube faces. Model (~6GB) and datasets released.
Closest neighbours by meaning, not just by category.
Model
Image-based model that predicts depth maps, surface normals and albedo from a single regular image.
Model
Research generative world model that creates consistent whole-house panorama tours from a floorplan and style reference.
Model
Research model that predicts layered, pixel-aligned 3D point clouds, including hidden geometry, from one image or video.
Research · Google DeepMind
Benchmark of perspective video clips cropped from 360° footage for evaluating camera pose estimation in SfM methods.
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.
