Symmetry-Aware 9D Pose Estimation with Sim(3)-Consistent Feature and Spherical Inception Convolution
About
Object pose estimation is a fundamental problem for an agent system to perceive or manipulate objects in images or videos. However, current instance-level methods struggle with generalization to unseen objects. Category-level methods seek to address this, but remain constrained by the complexities of learning in the non-linear Sim(3) space and intra-class variations. To address these challenges, We propose an effective method for category-level object pose estimation with two key innovations: (1) A translation/size estimator, featuring a semantic-guided symmetry-aware module that leverages robust generalization capabilities of a large vision model (LVM) to infer symmetry points, resulting in accurate translation and size without shape priors. This result serves as a precomputed cue for rotation estimation, thereby reducing the difficulty of learning in the non-linear Sim(3) space and laying a robust foundation for tackling the inherently more challenging rotation estimation. (2) A feature fusion module, based on our proposed spherical large-kernel inception convolution, fuses semantic features from the LVM with systematically computed geometric features to extract essential pose features from intra-class variations by modeling long-range dependencies without excessive computational cost. Built on these innovations, we achieve SOTA on benchmarks and real-world scenes, while developing a robust robotic picking system capable of handling diverse objects. Our code will be available at the project page: {\hypersetup{urlcolor=blue}https://panfei-cheng.github.io/SSH-Pose}.
Key insights
What problem does SSH-Pose solve?
It estimates 9D poses (translation, rotation, metric size) for novel objects within known categories without shape priors, tackling nonlinear Sim(3) space complexity and intra-class variation using symmetry-aware semantic features and spherical inception convolution.
How does it differ from prior work?
Unlike SecondPose's serial SE(3) pipeline, SSH-Pose uses Sim(3)-consistent geometric features enabling parallel processing, and replaces 3x3 convolutions with large-kernel spherical inception. On NOCS-REAL275 it exceeds SecondPose by 3.1 pp at 5°2cm mAP.
What practical use does this have?
The authors built a robotic picking prototype using a UR5 arm and dexterous hand, achieving 86% average success across five object categories and 100 trials, demonstrating real-world applicability beyond benchmark evaluation.
What limitations do the authors note?
Training requires scarce, hard-to-collect real-world data. The authors state future work will focus on training solely with synthetic datasets and breaking domain barriers to reduce training requirements.