OneCanvas: 3D Scene Understanding via Panoramic Reprojection

• URL: http://arxiv.org/abs/2606.19253v1
• Date: Wed, 17 Jun 2026 16:29:19 GMT
• Title: OneCanvas: 3D Scene Understanding via Panoramic Reprojection
• Abstract summary: OneCanvas aggregates patch features from all views onto a single panoramic canvas.<n>A 3D position embedding the patch's metric coordinates is added to its feature, restoring the depth lost when collapsing the world to angular canvas coordinate.<n>Because the canvas can be centered on any pose of interest, the same representation directly supports situated reasoning from a specific viewpoint.
• Score: 50.6678644849432
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Existing approaches to 3D scene understanding in Vision-Language Models (VLMs) either rely on complex, model-specific geometry encoders or large training budgets in pursuit of spatial reasoning. Instead, OneCanvas aggregates patch features from all views onto a single equirectangular panoramic canvas. Namely, each patch is unprojected to a 3D world coordinate using its depth and camera pose, then placed on the canvas at the continuous longitude and latitude of that point as seen from the canvas origin, with no rasterization or aggregation across overlapping views. A 3D position embedding of the patch's metric coordinates is added to its feature, restoring the depth lost when collapsing the world position to an angular canvas coordinate. Patches from all frames thus share one spatial coordinate system with no fusion or major architectural modifications of the backbone. The pretrained VLM consumes this representation as if it were an ordinary image. Because the canvas can be centered on any pose of interest, the same representation directly supports situated reasoning from a specific viewpoint, a common requirement in robotics and embodied AI. Thanks to this representation, we can also introduce a spatial pretraining curriculum: by procedurally placing patch features of objects, drawn from real images, at chosen 3D world positions on an otherwise empty canvas, we generate on-the-fly supervision spanning a broad range of spatial reasoning tasks, with answer distributions controlled to reduce spatial reasoning shortcuts. OneCanvas achieves state-of-the-art accuracy on SQA3D and VSI-Bench, and generalizes to out-of-distribution data on SPBench, using an order of magnitude less training compute than the strongest competing methods.

Related papers

• Sparse auto-regressive modeling for scene generation from multi-view images [20.272184280044083]
  We introduce SPAR3S, a sparse voxel-aligned 3D latent generative model for conditional scene completion.<n>Our key insight is to formulate 3D scene generation in a structured, compact, voxel-aligned 3D latent space.<n>We demonstrate the effectiveness of our method on synthetic indoor scenes, achieving higher novel-view quality than prior work.
  arXiv  Detail & Related papers  (2026-09-03T14:41:34Z)
• Dexterity-BEV: Aligning 3D World and Actions for Generalizable Robot Policies Learning [51.799524981291235]
  End-to-end manipulation policies show promise for generalizable and dexterous robotic manipulation.<n>They inherit two key limitations from 2D foundation models.<n>We present a series of contributions to address these issues.
  arXiv  Detail & Related papers  (2026-06-01T14:01:11Z)
• Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence [28.712695548782037]
  We present Tango3D, a foundation model that unifies dense correspondence and global retrieval.<n>We use a geometry-aware 2D visual backbone and a pretrained 3D VAE to encode images into 2D patches and point clouds into 3D tokens.<n>Experiments show our model successfully achieves object-level pixel-to-point alignment while maintaining competitive global retrieval.
  arXiv  Detail & Related papers  (2026-05-19T12:01:23Z)
• Geometry-Guided 3D Visual Token Pruning for Video-Language Models [35.786221225043455]
  We propose Geo3DPruner, a Geometry-Guided 3D Visual Token Pruning framework.<n>Geo3DPruner first models cross-frame relevance through geometry-aware global attention, and then performs a two-stage pruning process.<n>Experiments on multiple 3D scene understanding benchmarks demonstrate that Geo3DPruner retains over 90% of the original performance while pruning 90% of visual tokens, significantly outperforming existing text-guided and vision-guided pruning methods.
  arXiv  Detail & Related papers  (2026-04-20T13:33:50Z)
• OneWorld: Taming Scene Generation with 3D Unified Representation Autoencoder [90.8453349494245]
  We present OneWorld, a framework that performs diffusion directly within a coherent 3D representation space.<n>OneWorld generates high-quality 3D scenes with superior cross-view consistency compared to state-of-the-art 2D-based methods.
  arXiv  Detail & Related papers  (2026-03-17T03:43:37Z)
• Pano360: Perspective to Panoramic Vision with Geometric Consistency [7.713672589538202]
  We adopt a novel transformer-based architecture to achieve 3D awareness and aggregate global information across all views.<n>To establish an evaluation benchmark and train our network, we constructed a large-scale dataset of real-world scenes.
  arXiv  Detail & Related papers  (2026-03-12T14:56:14Z)
• 3D Congealing: 3D-Aware Image Alignment in the Wild [44.254247801001675]
  3D Congealing is a problem of 3D-aware alignment for 2D images capturing semantically similar objects. We introduce a general framework that tackles the task without assuming shape templates, poses, or any camera parameters. Our framework can be used for various tasks such as correspondence matching, pose estimation, and image editing.
  arXiv  Detail & Related papers  (2024-04-02T17:32:12Z)
• Generating Visual Spatial Description via Holistic 3D Scene Understanding [88.99773815159345]
  Visual spatial description (VSD) aims to generate texts that describe the spatial relations of the given objects within images. With an external 3D scene extractor, we obtain the 3D objects and scene features for input images. We construct a target object-centered 3D spatial scene graph (Go3D-S2G), such that we model the spatial semantics of target objects within the holistic 3D scenes.
  arXiv  Detail & Related papers  (2023-05-19T15:53:56Z)
• CAPE: Camera View Position Embedding for Multi-View 3D Object Detection [100.02565745233247]
  Current query-based methods rely on global 3D position embeddings to learn the geometric correspondence between images and 3D space. We propose a novel method based on CAmera view Position Embedding, called CAPE. CAPE achieves state-of-the-art performance (61.0% NDS and 52.5% mAP) among all LiDAR-free methods on nuScenes dataset.
  arXiv  Detail & Related papers  (2023-03-17T18:59:54Z)
• Sparse Pose Trajectory Completion [87.31270669154452]
  We propose a method to learn, even using a dataset where objects appear only in sparsely sampled views. This is achieved with a cross-modal pose trajectory transfer mechanism. Our method is evaluated on the Pix3D and ShapeNet datasets.
  arXiv  Detail & Related papers  (2021-05-01T00:07:21Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.