Semantic Prior Guided One-View 6D Pose Estimation for Novel Objects

2026-07-23 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

OneViewAll is a novel framework for 6D object pose estimation designed for novel objects, operating efficiently with only a single real-world RGB-D reference view and no explicit CAD models. It introduces a "Project-and-Compare" paradigm, directly aligning reference and query observations within a projection-equivariant space, bypassing computationally expensive CAD-based rendering. The system integrates hierarchical semantic priors across three levels: category- and scene-level for hypothesis initialization, object-level symmetry for geometry completion via mirror fusion, and patch-level for discriminative refinement. Extensive experiments demonstrate OneViewAll achieves 92.5% ADD-0.1 accuracy on the LINEMOD dataset, significantly outperforming the CVPR 2025 baseline One2Any (52.6%). It also shows consistent improvements on YCB-V, Real275, and Toyota-Light, maintaining low inference latency, such as 80 ms for 12 hypotheses on an NVIDIA RTX 4090 GPU.

Key takeaway

For Robotics Engineers developing systems for dexterous manipulation or augmented reality with novel objects, OneViewAll offers a robust and efficient 6D pose estimation solution. Its ability to achieve high accuracy (92.5% ADD-0.1 on LINEMOD) from a single RGB-D reference view, without CAD models, significantly reduces data acquisition and computational overhead. Consider integrating this "Project-and-Compare" framework to improve real-time performance and generalization across unseen objects in cluttered environments.

Key insights

OneViewAll enables accurate, efficient 6D pose estimation for novel objects using a single RGB-D view and hierarchical semantic priors.

Principles

• Direct observation alignment avoids CAD rendering overhead.
• Symmetry priors complete invisible object geometry.
• Hierarchical semantics guide coarse-to-fine pose refinement.

Method

OneViewAll initializes pose hypotheses using semantic priors, iteratively refines them via symmetry-aware projection and patch-wise attention, then selects the optimal pose.

In practice

• Use single RGB-D views for novel object pose tasks.
• Apply symmetry-aware projection for occluded objects.
• Integrate hierarchical priors for robust pose estimation.

Topics

• 6D Pose Estimation
• Model-Free Pose Estimation
• RGB-D Vision
• Semantic Priors
• Symmetry-Aware Geometry
• Robotic Perception

Code references

• tilaba/OneViewAll

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• One-Shot Novel View and Pose Human Image Synthesis via 3D Prior Guided Diffusion Model — A 3D prior-guided conditional diffusion model enables high-quality one-shot novel view and pose human image synthesis.
• Rotational Symmetry based Object Pose Estimation from Point Clouds in the Absence of Known 3D Models — Leveraging rotational symmetry enables robust object pose estimation from point clouds even without known 3D models, using iterative optimization.
• Every9D-21M: Large-Scale Real-World 9D Canonicalization of Everyday Objects — Large-scale, real-world 9D pose datasets can be built efficiently by propagating sparse manual annotations across object-centric videos.
• GAP-GDRNet: Geometry-Aware Monocular Visual Pose Sensing on a Single-Target Synthetic Spacecraft Dataset — GAP-GDRNet enhances 6D spacecraft pose sensing by integrating attention mechanisms for robust geometric feature extraction.
• Unsupervised Pixel-Level Semantic Left-Right Understanding of In-the-Wild Images — An unsupervised framework predicts pixel-level semantic left-right understanding in diverse images by combining 3D shape and image datasets.
• GARDEN: Gravity-Aligned Reconstruction of Disentangled ENvironments from RGB images — Gravity provides a universal physical prior for robust, physically-grounded 3D scene reconstruction from RGB images.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.