Seeking Consensus: Geometric-Semantic On-the-Fly Recalibration for Open-Vocabulary Remote Sensing Semantic Segmentation

2026-04-28

AI Summary

Existing open-vocabulary semantic segmentation (OVSS) methods for remote sensing rely on static inference, which struggles to adapt to scene-specific characteristics and often suffers from semantic ambiguity and insufficient foreground activation. To address these limitations, this work proposes SeeCo, a novel framework that introduces a dual-consensus mechanism—geometric consensus learning grounded in multi-view consistency and semantic consensus learning driven by adaptive calibration via textual descriptions. This mechanism enables plug-and-play, scene-level dynamic recalibration during inference for any OVSS model without requiring additional training. By mitigating under-activation and bias issues in vision–language alignment, SeeCo consistently improves performance across eight remote sensing OVSS benchmarks, demonstrating its effectiveness and broad applicability.

Abstract

Open-vocabulary semantic segmentation (OVSS) in remote sensing images is a promising task that employs textual descriptions for identifying undefined land cover categories. Despite notable advances, existing methods typically employ a static inference paradigm, overlooking the distinct distribution of each scene, resulting in semantic ambiguity in diverse land covers and incomplete foreground activation. Motivated by this, we propose Seeking Consensus, termed SeeCo, a plug-and-play framework to boost the performance of training-free OVSS models in remote sensing images, which recalibrates arbitrary OVSS models on-the-fly by seeking dual consensus: geometric consensus learning (GCL) through multi-view consistent observations and semantic consensus learning (SCL) via textual description adaptive calibration, which assists collaborative recalibration of visual and textual semantics. The two consensus are injected via an online consensus injector (OCI), effectively alleviating the under-activation and semantic bias. SeeCo requires no specific training process, yet recalibrates semantic-geometric alignment for each unique scene during inference. Extensive experiments on eight remote sensing OVSS benchmarks show consistent gains, proving its effectiveness and universality.

Problem

• open-vocabulary semantic segmentation
• remote sensing
• semantic ambiguity
• foreground activation
• scene-specific distribution

Innovation

• open-vocabulary segmentation
• geometric consensus
• semantic recalibration
• remote sensing
• training-free adaptation