Global Truncated Loss Minimization for Robust and Threshold-Resilient Geometric Estimation
Tianyu Huang, Liangzu Peng, Xinyue Zhang, Tongfan Guan, Jinhu Dong, Haoang Li · 2026
Read Global Truncated Loss Minimization for Robust and Threshold-Resilient Geometric Estimation by Tianyu Huang, Liangzu Peng, Xinyue Zhang, Tongfan Guan, Jinhu Dong, Haoang Li (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14796.txt.
Introduction
Huang and co-authors minimize truncated losses globally for robust geometric estimation with GTM. Consensus maximization resists outliers but leans on inlier counts alone. Its score jumps with the inlier threshold, and bounds stay loose. Truncated losses use residual size, a richer signal. GTM is the first branch-and-bound frame to minimize them globally. Search covers all but one dimension, with the last solved by bounding. Lipschitz bounds that are tight and general speed the hunt. Threshold resilience and search gains span many geometric tasks, they report.
Catalog record
| CREATOR | Tianyu Huang, Liangzu Peng, Xinyue Zhang, Tongfan Guan, Jinhu Dong, Haoang Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 104290 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.14796
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Global Truncated Loss Minimization for Robust and Threshold-Resilient Geometric Estimation Thanks: T. Huang, T. Guan, J. Dong, and Y.-H. Liu are with the Hong Kong Centre For Logistics Robotics, The Chinese University of Hong Kong. L. Peng is with the Center for Innovation in Data Engineering and Science (IDEAS), University of Pennsylvania. X. Zhang and L. Kneip are with the Mobile Perception Lab of the School of Information and Technology, ShanghaiTech University. H. Li is with the Thrust of Robotics and Autonomous Systems, The Hong Kong University of Science and Technology (Guangzhou). Yun-Hui Liu is the corresponding author. E-mail: [email protected] Tianyu Huang, Liangzu Peng, Xinyue Zhang, Tongfan Guan, Jinhu Dong, Haoang Li Abstract To achieve outlier-robust geometric estimation, robust objective functions are generally employed to mitigate the influence of outliers. The widely used consensus maximization (CM) is highly robust when paired with global branch-and-bound (BnB) search. However, CM relies solely on inlier counts and is sensitive to the inlier threshold. Besides, the discrete nature of CM leads to loose bounds, necessitating extensive BnB iterations and computation cost. Truncated losses (TL), another continuous alternative, leverage residual information more effectively and could potentially overcome these issues. But to our knowledge, no prior work has systematically explored globally minimizing TL with BnB and its potential for enhanced threshold resilience or search efficiency. In this work, we propose GTM , the first unified BnB-based framework for g lobally-optimal T L loss m inimization across diverse geometric problems. GTM involves a hybrid solving design: given an n n -dimensional problem, it performs BnB search over an ( n − 1 ) (n-1) -dimensional subspace while the remaining 1D variable is solved by analytically bounding the objective function. Our hybrid design not only reduces the search space, but also enables us to derive Lipschitz-continuous bounding functions that are general, tight, and can be efficiently solved by a classic global Lipschitz solver named DIRECT, which brings further acceleration. We conduct a systematic evaluation on the performance of various BnB-based methods for CM and TL on the robust linear regression problem, showing that GTM enjoys remarkable threshold resilience and the highest efficiency compared to baseline methods. Furthermore, we apply GTM on different geometric estimation problems with diverse residual forms. Both simulated and real-world experiments demonstrate that GTM achieves state-of-the-art outlier-robustness and threshold-resilience while maintaining high efficiency across these estimation tasks. I Introduction Geometric estimation is fundamental to numerous computer vision and robotics applications, e.g., scene reconstruction [ 38 , 70 ] , robot navigation [ 29 , 65 ] , panoramic stitching [ 13 , 14 ] , etc. Its core objective is fitting geometric models (e.g., camera poses) from observed data that are often represented as feature matches [ 2 , 20 ] . Despite the importance, this task is challenging due to the presence of outlier data caused by noisy measurement and imperfect feature description. This challenge motivates the development of outlier-robust estimation techniques, particularly those grounded in robust objective functions. One prominent robust objective function is consensus maximization (CM) [ 27 , 22 , 35 , 36 ] . For a specific estimation problem, we often have a residual function r:\mathcal{C}\times\mathcal{D}\rightarrow[0,\infty) (e.g., re-projection error), where \mathcal{C}\subset\mathbb{R}^{n} defines the feasible region of the geometric variable \mathbf{v} and \mathcal{D} is the data domain. Given measured data \{\mathbf{d}_{i}\}_{i=1}^{M}\subset\mathcal{D} and a predefined threshold \xi , CM aims to find a \mathbf{v}\in\mathcal{C} that maximizes the number of inlier data \mathbf{d}_{i} satisfying r(\mathbf{v},\mathbf{d}_{i})\leq\xi , i.e., \displaystyle\max_{\mathbf{v}\in\mathcal{C}}\sum_{i=1}^{M}\bm{1}(r(\mathbf{v},\mathbf{d}_{i})\leq\xi), ( CM ) where \bm{1}(\cdot) outputs 1 if the input statement is true, or otherwise outputs 0 . Despite its nonconvexity, ( CM ) has been addressed by various powerful strategies [ 69 , 5 , 32 ] , particularly the branch-and-bound (BnB) method that can provide globally optimal solutions through systematic search [ 36 , 35 , 18 , 16 ] . However, the CM objective in ( CM ) suffers from inherent limitations. First, it relies solely on inlier counts and is sensitive to threshold \xi (see Fig. 1 ). In practice, selecting a suitable threshold is an art and one cannot choose it effectively without any prior knowledge [ 33 , 26 ] . Second, the discrete nature of ( CM ) complicates the development of tight bounds for BnB, often leading to extensive iterations for convergence [ 37 , 77 ] . On the other hand, the truncated losses (TL), another classical robust objective function [ 37 , 39 , 11 , 2 ] , not only mitigates the influence of outliers with residuals higher than a threshold as in ( CM ), but also goes further to minimize the residuals below this threshold: \min_{\mathbf{v}\in\mathcal{C}}\sum_{i=1}^{M}\min\{r(\mathbf{v},\mathbf{d}_{i}),\ \xi\}. ( TL ) Would the continuous objective in ( TL ) overcome the above limitations of ( CM ), particularly the sensitivity to misspecified thresholds? Since, to the best of our knowledge, no prior work proves this intuition, in Fig. 1 we conduct a simple experiment to verify it and observe: as \xi increases, the ground truth remains the global optimizer of ( TL ) but no longer maximizes ( CM ), highlighting the superior threshold-resilience of ( TL ). Having verified the advantage of ( TL ), the next critical question arises: Can we solve ( TL ) to global optimality? In contrast to the established success of solving ( CM ) via global optimization techniques like BnB, TL has received significantly less exploration in this regard. Indeed, most existing solvers do not guarantee solving ( TL ) global optimally [ 11 , 61 , 2 , 71 ] . To the best of our knowledge, the only attempt to solve ( TL ) optimally is [ 37 ] , but the formulation and BnB algorithm of [ 37 ] are tailored to a specific geometric problem with a specific residual function (in the context of point cloud registration ), thus it is not applicable to ( TL ) with diverse residuals. In this work, we introduce the first unified BnB-based framework designed to solve ( TL ) across a diverse range of geometric tasks characterized by various residual functions, named G lobally-optimal T L loss M inimization ( GTM ). At the heart of GTM is a strategic blend of two powerful and complementary global optimization mechanisms. Instead of conducting a conventional BnB search over the full n -dimensional variable space, GTM confines the BnB procedure to an (n-1) -dimensional subspace . Concurrently, the remaining 1-dimensional variable is addressed using the DIRECT method [ 42 , 41 ] , a derivative-free optimization strategy renowned for its efficiency in optimizing Lipschitz continuous functions. The rationale of employing the ( n-1 )-dimensional BnB is that it reduces the search dimension by one, leading to appreciable acceleration as evidenced in different contexts from prior studies [ 59 , 77 , 37 ] . The rationale of employing the DIRECT method is that it perfectly addresses two fundamental challenges that arise from ( n-1 )-dimensional BnB: (1) the need to derive general bounding functions for diverse residuals; (2) globally optimizing the bounding functions efficiently. To deal with the first challenge, we leverage interval arithmetic [ 31 , 55 ] to derive innovative general bounding functions that require