Depth Anything in $360^\circ$: Towards Scale Invariance in the Wild
Read the original on arXiv Computer Vision →The Flow has not summarised this story yet — read it at arXiv Computer Vision.
The Flow has not summarised this story yet — read it at arXiv Computer Vision.
The paper introduces PaGeR, a framework that adapts pre‑trained 3D reconstruction transformers to handle both perspective and panoramic images. PaGeR predicts scale‑invariant depth, metric depth, surface normals, and sky masks in a single forward pass, using minimal architectural changes and mixed‑domain training. Extensive indoor and outdoor tests show state‑of‑the‑art and strong zero‑shot performance across diverse scenes.
arXiv:2610.19502v1 Announce Type: new Abstract: Spatial intelligence requires scalable 3D environments that are both photorealistic and semantically grounded, yet existing indoor datasets are dominat...
The paper introduces PVDepth, an end‑to‑end framework for panoramic video depth estimation that adapts perspective video depth models to the equirectangular projection (ERP) domain. It first presents PanoCARLA, a large‑scale synthetic RGB‑D panoramic video dataset with natural motion trajectories and drone‑like viewpoints. PVDepth addresses ERP‑specific challenges with a Progressive Sphere‑aware Noise Initialization strategy and a Cube‑rectified Temporal Modeling module, achieving superior geometrically accurate and temporally consistent depth sequences.
PanoWorld is a panoramic video world model that generates geometry‑consistent 360° video from a single image and a caption. It frames panoramic video generation as a geometry‑ and dynamics‑consistent latent state modeling problem, using depth consistency and trajectory consistency regularizers built on a pre‑trained perspective video world model. The authors also introduce PanoGeo, a unified geometry‑aware panoramic video dataset with consistent depth, trajectory, and prompt annotations, and demonstrate that PanoWorld improves geometric consistency while maintaining competitive visual realism.
Dyna3 is a training‑free framework that extends the depth foundation model DA3 to perform 4D dynamic scene reconstruction without fine‑tuning. By leveraging DA3’s cross‑view features and a best‑match search, it distinguishes static surfaces from moving objects, and uses vision‑language models to generate semantic prompts for SAM 3 to achieve precise instance‑level segmentation. Experiments on four datasets show Dyna3 outperforms correspondence‑trained methods, improving dynamic object segmentation by +5.5 pp, speeding pose estimation 13×, and reducing memory usage 4–8×.
The paper introduces Self-Geometry, a plug‑and‑play test‑time adaptation framework that enforces explicit multi‑view geometric constraints on Vision Foundation Models (VFMs) using 2D pixel correspondences as pseudo ground truth. It combines Geometric Disentanglement Optimization—mixing Multi‑View and Epipolar Consistency losses with Gradient Disentanglement—to avoid gradient conflicts, a Frame Angular‑Neighbor sampler based on SO(3) geodesic distances to select informative views, and a Lightweight TTA module that adapts VFMs via LoRA. Experiments on six VFMs and four benchmarks (7Scenes, ETH3D, ScanNet++, HiRoom) show consistent improvements in pose and geometry estimation.