DecomPose: Disentangling Cross-Category Optimization Contention for Category-Level 6D Object Pose Estimation
The paper titled 'DecomPose' addresses challenges in category-level 6D object pose estimation caused by cross-category optimization contention. It introduces a framework that utilizes difficulty-aware gradient decoupling and stability-driven asymmetric branching to improve pose estimation performance. Extensive experiments demonstrate the effectiveness of DecomPose in reducing optimization contention across various benchmarks.
- ▪The study focuses on category-level 6D object pose estimation as a multi-category joint learning problem.
- ▪It identifies that geometric heterogeneity across categories leads to gradient conflicts during training.
- ▪The proposed DecomPose framework mitigates these conflicts through difficulty-aware gradient decoupling and stability-driven asymmetric branching.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.15728 (cs) [Submitted on 15 May 2026] Title:DecomPose: Disentangling Cross-Category Optimization Contention for Category-Level 6D Object Pose Estimation Authors:Yifan Gao, Lu Zou, Zhangjin Huang, Guoping Wang View a PDF of the paper titled DecomPose: Disentangling Cross-Category Optimization Contention for Category-Level 6D Object Pose Estimation, by Yifan Gao and 2 other authors View PDF HTML (experimental) Abstract:Category-level 6D object pose estimation is typically formulated as a multi-category joint learning problem with fully shared model parameters.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
