CoRegOVCD: Consistency-Regularized Open-Vocabulary Change Detection

2026-04-02

AI Summary

Existing remote sensing change detection methods are constrained by fixed label spaces, limiting their ability to support arbitrary semantic queries, while training-free open-vocabulary approaches often suffer from unstable cross-temporal concept responses and spatial fragmentation, leading to suboptimal performance. This work proposes a training-free dense inference framework that, for the first time, integrates posterior calibration with geometric structural consistency. It models semantic change through Competitive Posterior Calibration (CPC) and Semantic Posterior Difference (SPD), and enhances spatial coherence via a Geometry-to-Token Consistency Gating mechanism (GeoGate) and Region Consensus Discrepancy (RCD). The proposed method significantly outperforms the strongest training-free baselines across four benchmark datasets, achieving F1_C improvements of 2.24–4.98 points and attaining an average F1_C of 47.50% across six classes on the SECOND dataset.

Abstract

Remote sensing change detection (CD) aims to identify where land-cover semantics change across time, but most existing methods still assume a fixed label space and therefore cannot answer arbitrary user-defined queries. Open-vocabulary change detection (OVCD) instead asks for the change mask of a queried concept. In the fully training-free setting, however, dense concept responses are difficult to compare directly across dates: appearance variation, weak cross-concept competition, and the spatial continuity of many land-cover categories often produce noisy, fragmented, and semantically unreliable change evidence. We propose Consistency-Regularized Open-Vocabulary Change Detection (CoRegOVCD), a training-free dense inference framework that reformulates concept-specific change as calibrated posterior discrepancy. Competitive Posterior Calibration (CPC) and the Semantic Posterior Delta (SPD) convert raw concept responses into competition-aware queried-concept posteriors and quantify their cross-temporal discrepancy, making semantic change evidence more comparable without explicit instance matching. Geometry-Token Consistency Gate (GeoGate) and Regional Consensus Discrepancy (RCD) further suppress unsupported responses and improve spatial coherence through geometry-aware structural verification and regional consensus. Across four benchmarks spanning building-oriented and multi-class settings, CoRegOVCD consistently improves over the strongest previous training-free baseline by 2.24 to 4.98 F1$_C$ points and reaches a six-class average of 47.50% F1$_C$ on SECOND.

Problem

• open-vocabulary change detection
• remote sensing
• semantic change
• training-free
• concept response

Innovation

• open-vocabulary change detection
• posterior calibration
• spatial consistency
• training-free inference
• semantic discrepancy