Robust 4D Visual Geometry Transformer with Uncertainty-Aware Priors

2026-04-10

AI Summary

This work addresses the challenge of geometric blur induced by motion in dynamic 4D scene reconstruction, where existing 3D foundation models struggle to effectively represent dynamic sequences. The authors propose an uncertainty-aware decoupled reconstruction framework that explicitly separates dynamic and static components through three key innovations: entropy-guided subspace projection, local consistency-based geometric purification, and uncertainty-aware cross-view consistency. By integrating information-theoretic weighted multi-head attention, radius-neighborhood spatial constraints, and heteroscedastic maximum likelihood estimation, the method achieves superior performance without requiring per-scene fine-tuning or optimization. On dynamic benchmarks, it reduces average geometric error by 13.43% and improves segmentation F-score by 10.49%, significantly outperforming current state-of-the-art approaches.

Technology Category

• Computer Vision: Motion & Tracking
• Reasoning under Uncertainty: Uncertainty Representations
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Reconstructing dynamic 4D scenes is an important yet challenging task. While 3D foundation models like VGGT excel in static settings, they often struggle with dynamic sequences where motion causes significant geometric ambiguity. To address this, we present a framework designed to disentangle dynamic and static components by modeling uncertainty across different stages of the reconstruction process. Our approach introduces three synergistic mechanisms: (1) Entropy-Guided Subspace Projection, which leverages information-theoretic weighting to adaptively aggregate multi-head attention distributions, effectively isolating dynamic motion cues from semantic noise; (2) Local-Consistency Driven Geometry Purification, which enforces spatial continuity via radius-based neighborhood constraints to eliminate structural outliers; and (3) Uncertainty-Aware Cross-View Consistency, which formulates multi-view projection refinement as a heteroscedastic maximum likelihood estimation problem, utilizing depth confidence as a probabilistic weight. Experiments on dynamic benchmarks show that our approach outperforms current state-of-the-art methods, reducing Mean Accuracy error by 13.43\% and improving segmentation F-measure by 10.49\%. Our framework maintains the efficiency of feed-forward inference and requires no task-specific fine-tuning or per-scene optimization.

Problem

• 4D reconstruction
• dynamic scenes
• geometric ambiguity
• motion cues
• uncertainty modeling

Innovation

• Uncertainty-Aware Priors
• Entropy-Guided Subspace Projection
• Local-Consistency Geometry Purification
• Heteroscedastic Maximum Likelihood
• 4D Scene Reconstruction