ChronoGS: Disentangling Invariants and Changes in Multi-Period Scenes

2025-11-24

AI Summary

Long-term, non-continuous geometric and appearance changes in multi-epoch scenarios (e.g., urban mapping, construction site monitoring) violate the static or dynamic scene assumptions underlying existing 3D reconstruction methods, leading to failure. To address this, we propose Temporal-Modulated Gaussians (TM-Gaussians), a unified anchor-based representation enabling consistent cross-epoch modeling. TM-Gaussians are the first method to explicitly decouple stable and evolving scene components, overcoming the modeling bottleneck under non-continuous temporal changes. Our approach integrates temporal modulation, differentiable rendering, and joint spatio-temporal optimization, trained and validated on our newly introduced ChronoScene dataset—a large-scale real-world and synthetic benchmark for time-varying scenes. Extensive experiments demonstrate significant improvements in both reconstruction accuracy and temporal consistency over diverse baselines. Code and the ChronoScene dataset are publicly released.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Multi-modal Vision
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Multi-period image collections are common in real-world applications. Cities are re-scanned for mapping, construction sites are revisited for progress tracking, and natural regions are monitored for environmental change. Such data form multi-period scenes, where geometry and appearance evolve. Reconstructing such scenes is an important yet underexplored problem. Existing pipelines rely on incompatible assumptions: static and in-the-wild methods enforce a single geometry, while dynamic ones assume smooth motion, both failing under long-term, discontinuous changes. To solve this problem, we introduce ChronoGS, a temporally modulated Gaussian representation that reconstructs all periods within a unified anchor scaffold. It's also designed to disentangle stable and evolving components, achieving temporally consistent reconstruction of multi-period scenes. To catalyze relevant research, we release ChronoScene dataset, a benchmark of real and synthetic multi-period scenes, capturing geometric and appearance variation. Experiments demonstrate that ChronoGS consistently outperforms baselines in reconstruction quality and temporal consistency. Our code and the ChronoScene dataset are publicly available at https://github.com/ZhongtaoWang/ChronoGS.

Problem

• Reconstructing multi-period scenes with evolving geometry and appearance
• Disentangling stable and changing components across time periods
• Overcoming limitations of static and dynamic reconstruction methods

Innovation

• Temporally modulated Gaussian representation for scenes
• Unified anchor scaffold reconstructs all periods
• Disentangles stable and evolving scene components