Tail-Aware Post-Training Quantization for 3D Geometry Models

2026-02-02

AI Summary

This work addresses the poor performance of conventional post-training quantization (PTQ) on 3D geometric models, which stems from complex feature distributions and high calibration costs. To overcome these limitations, the authors propose TAPTQ, a tail-aware PTQ framework that integrates progressive coarse-to-fine calibration, an efficient quantization range optimization based on ternary search, and a module-level compensation mechanism guided by Tail Relative Error (TRE). This approach simultaneously enhances quantization accuracy and substantially reduces calibration overhead. Experimental results demonstrate that TAPTQ significantly outperforms existing PTQ methods on both VGGT and Pi3 benchmarks, achieving superior accuracy with notably lower computational cost.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: 3D Computer Vision
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The burgeoning complexity and scale of 3D geometry models pose significant challenges for deployment on resource-constrained platforms. While Post-Training Quantization (PTQ) enables efficient inference without retraining, conventional methods, primarily optimized for 2D Vision Transformers, fail to transfer effectively to 3D models due to intricate feature distributions and prohibitive calibration overhead. To address these challenges, we propose TAPTQ, a Tail-Aware Post-Training Quantization pipeline specifically engineered for 3D geometric learning. Our contribution is threefold: (1) To overcome the data-scale bottleneck in 3D datasets, we develop a progressive coarse-to-fine calibration construction strategy that constructs a highly compact subset to achieve both statistical purity and geometric representativeness. (2) We reformulate the quantization interval search as an optimization problem and introduce a ternary-search-based solver, reducing the computational complexity from $\mathcal{O}(N)$ to $\mathcal{O}(\log N)$ for accelerated deployment. (3) To mitigate quantization error accumulation, we propose TRE-Guided Module-wise Compensation, which utilizes a Tail Relative Error (TRE) metric to adaptively identify and rectify distortions in modules sensitive to long-tailed activation outliers. Extensive experiments on the VGGT and Pi3 benchmarks demonstrate that TAPTQ consistently outperforms state-of-the-art PTQ methods in accuracy while significantly reducing calibration time. The code will be released soon.

Problem

• Post-Training Quantization
• 3D Geometry Models
• Feature Distribution
• Calibration Overhead
• Long-tailed Outliers

Innovation

• Post-Training Quantization
• 3D Geometry Models
• Tail-Aware Quantization
• Ternary Search Optimization
• Calibration Subset Construction