Be the Change You Want to See: Revisiting Remote Sensing Change Detection Practices

2025-07-04

AI Summary

Remote sensing change detection research often emphasizes designing complex novel modules while overlooking the substantial impact of fundamental design choices—such as backbone architectures, pretraining strategies, and training configurations. This work systematically revisits the design space of change detection through rigorous ablation studies and comprehensive evaluation across six benchmark datasets, revealing that optimizing foundational components yields greater performance gains than structural innovation alone. We propose a set of generalizable, principle-driven guidelines for foundational optimization and construct a lightweight, efficient baseline framework. Critically, this baseline introduces no additional parameters or computational overhead, yet achieves state-of-the-art (SOTA) or superior performance on all six datasets. Moreover, our optimization strategy exhibits strong transferability: when applied to diverse existing methods, it consistently delivers significant performance improvements. These findings establish a new paradigm for change detection model design—one grounded in principled, systematic refinement of core architectural and training decisions rather than incremental module engineering.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Other Foundations of Computer Vision
• Search and Optimization: Other Foundations of Search & Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Remote sensing change detection aims to localize semantic changes between images of the same location captured at different times. In the past few years, newer methods have attributed enhanced performance to the additions of new and complex components to existing architectures. Most fail to measure the performance contribution of fundamental design choices such as backbone selection, pre-training strategies, and training configurations. We claim that such fundamental design choices often improve performance even more significantly than the addition of new architectural components. Due to that, we systematically revisit the design space of change detection models and analyse the full potential of a well-optimised baseline. We identify a set of fundamental design choices that benefit both new and existing architectures. Leveraging this insight, we demonstrate that when carefully designed, even an architecturally simple model can match or surpass state-of-the-art performance on six challenging change detection datasets. Our best practices generalise beyond our architecture and also offer performance improvements when applied to related methods, indicating that the space of fundamental design choices has been underexplored. Our guidelines and architecture provide a strong foundation for future methods, emphasizing that optimizing core components is just as important as architectural novelty in advancing change detection performance. Code: https://github.com/blaz-r/BTC-change-detection

Problem

• Evaluates impact of backbone selection and pre-training on change detection
• Optimizes core design choices over adding complex architectural components
• Demonstrates simple models can outperform state-of-the-art with proper tuning

Innovation

• Revisits fundamental design choices for change detection
• Optimizes backbone selection and pre-training strategies
• Simplifies architecture while matching state-of-the-art performance