TriCons-Pose Triangle-Invariant Geometric Consistency Learning for Category-Level Object Pose Estimation

Jul 16, 2026

Upload by Rowan

Category-level object pose estimation is a key yet difficult task in both research and industry, especially when relying on keypoint-based correspondence. Many approaches strengthen feature learning while failing to ensure that correspondences remain geometrically stable under diverse perturbations, causing fragile pose recovery under intra-class shape variations and occlusions. TriCons-Pose introduces a triangle-invariant geometric consistency learning scheme that anchors stable keypoints and aggregates pose-invariant cues. It employs an SCKD with cross-view normalized pairwise distance matching, and a PIGA using triangle-based descriptors within local-to-global attention, optimized with an added geometry consistency loss. Experiments on REAL275, CAMERA25, and HouseCat6D validate effectiveness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

Introduction

Background and problem definition

Prior-based vs prior-free methods

Keypoint-based correspondence and geometric instability

TriCons-Pose: Triangle-Invariant Geometric Consistency Learning for Category-Level Object Pose Estimation Zuzhi Yang, Shuai Wang, Mounir Kaaniche, Senior Member, IEEE, Ziwei Li, Zhiming Cheng, Zhidong Zhao, Chenggang Yan arXiv :2607 . 10754v1 [ cs .CV] 12 Jul 2026 Abstract—Category-level object pose estimation is a crucial yet challenging task in both academia and industry, and has achieved remarkable success by leveraging keypoint-based correspondence paradigms. However, most existing methods increasingly rely on stronger feature learning while overlooking whether the established correspondences are geometrically stable across diverse perturbations. This often results in fragile pose recovery under intra-class shape variations and occlusions. To tackle this challenge, we develop a novel Triangle-Invariant Geometric Consistency Learning for Category-Level Object Pose Estimation (TriCons-Pose) to anchor stable keypoints and aggregate poseinvariant cues, yielding reliable canonical mapping and accurate pose estimation. Specifically, a Structure-Consistent Keypoint Detector (SCKD) is designed to identify robust keypoints by enforcing cross-view structural consistency via normalized pairwise distance matching. Moreover, we propose a Pose-Invariant Geometric Aggregator (PIGA) to augment keypoint representations by injecting triangle-based pose-invariant descriptors into a localto-global attention mechanism. The proposed framework is optimized using standard objective functions while incorporating an additional geometry consistency loss. Extensive experiments on REAL275, CAMERA25, and HouseCat6D datasets demonstrate the effectiveness of the proposed approach. Index Terms—Category-level pose estimation, keypoint-based correspondence, structural and geometric consistency I. INTRODUCTION OBJECT pose estimation has received considerable atten tion in both academia and industry because of its broad applicability in robotic grasping [1], autonomous driving [2], augmented reality [3], and 3D scene understanding [4] . As a fundamental problem in 3D vision, 6D object pose estimation seeks to determine the 3D rotation, 3D translation of an object, which are essential for geometric perception and subsequent physical interaction. With the support of instancespecific CAD models, instance-level object pose estimation [5] Zuzhi Yang, Shuai Wang, and Zhidong Zhao are with the School of Cyberspace, Hangzhou Dianzi University, Hangzhou 310018, China (e-mail: [242270060@hdu.edu.cn](242270060@hdu.edu.cn), [shuaiwang.tai@gmail.com](shuaiwang.tai@gmail.com), [zhaozd@hdu.edu.cn](zhaozd@hdu.edu.cn)). Zhiming Cheng and Chenggang Yan are with the School of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China (e-mail: [czming@hdu.edu.cn](czming@hdu.edu.cn), [cgyan@hdu.edu.cn](cgyan@hdu.edu.cn)). Mounir Kaaniche is with Universit Sorbonne Paris Nord, L2TI, UR 3043, F-93430, Villetaneuse, France, and Universit Paris-Saclay, CentraleSuplec, CVN, 91190 Gif-sur-Yvette, France (e-mail: mounir.kaaniche@univ[paris13.fr](paris13.fr)). Ziwei Li is with the School of Computer Science and Technology, University of Science and Technology of China, Hefei, China. (e-mail: [ziwei.li@kaust.edu.sa](ziwei.li@kaust.edu.sa)) . has achieved remarkable success, yielding accurate pose predictions for seen object instances. However, its dependence on predefined object models restricts applicability to objects without corresponding CAD models and limits generalization to unseen instances. To overcome these limitations, categorylevel object pose estimation [6], [7] seeks to estimate the pose and size of unseen objects from categories observed during training without requiring CAD models during inference. Despite recent progress, this task remains highly challenging due to intra-class shape variation, occlusion, and sensor noise. Existing category-level methods can generally be divided into prior-based and prior-free approaches. Pri

FAQs

What problem does TriCons-Pose address in category-level object pose estimation?

It addresses geometric instability of keypoint correspondences, which can drift across viewpoints and perturbations, leading to fragile pose recovery under intra-class shape variation and occlusion.

How does TriCons-Pose improve keypoint reliability?

It introduces a Structure-Consistent Keypoint Detector (SCKD) that enforces cross-view structural consistency through normalized pairwise distance matching to identify robust keypoints.

How are pose-invariant cues aggregated in the proposed framework?

It uses a Pose-Invariant Geometric Aggregator (PIGA) that injects triangle-based pose-invariant descriptors into a local-to-global attention mechanism to augment keypoint representations.

Jul 16, 2026

Upload by Rowan