CROSS is proposed, a tightly integrated paradigm for RRSIS that achieves state-of-the-art performance and maintains precise localization even under severe spatial description perturbations, standing as a robust new paradigm for RRSIS.
Abstract
Referring Remote Sensing Image Segmentation (RRSIS) has achieved significant progress through the integration of VLMs and the Segment Anything Model (SAM). However, this progress largely relies on strong pre-trained capabilities, while leaving two fundamental limitations insufficiently addressed: (1) Architectural Weak-Coupling, where the unidirectional flow forces reliance on coarse VLM prompts and wastes SAM's pixel-level structural guidance, causing localization drift; and (2) Object-Centric Semantic Bias, where models overemphasize dominant object semantics while remaining insensitive to spatial reasoning crucial for RRSIS. Motivated by these observations, we propose CROSS, a tightly integrated paradigm for RRSIS. First, we introduce Linguistic-Guided Cascaded Distillation (LGCD) to bridge the architectural gap, which distills SAM's geometric affinities as soft regularizers into VLM intermediate layers, injecting dense structural priors to refine localization. Second, Perspective-Spatial Contrastive Learning (PSCL) imposes cross-anchored constraints by mining mask-filtered deceptive distractors and spatial-linguistic counterfactuals as hard negatives, explicitly shattering semantic shortcuts to enforce genuine logical consistency. Extensive experiments on RRSIS benchmarks demonstrate that CROSS achieves state-of-the-art performance and maintains precise localization even under severe spatial description perturbations, standing as a robust new paradigm for RRSIS.
Referring remote sensing image segmentation (RRSIS) aims to segment target ground objects in high-resolution remote sensing imagery according to textual descriptions. Due to complex backgrounds and large-scale variations, accurately aligning linguistic semantics with spatial regions remains a challenging problem. Existing RRSIS methods are predominantly segmentation-centric, where textual descriptions mainly serve as conditional guidance for pixel-wise prediction rather than explicitly enforcing semantic consistency between visual regions and linguistic representations. This limitation often leads to suboptimal fine-grained alignment and insufficient suppression of background-dominant regions. To address these challenges, we propose a novel RRSIS framework centered on structure-constrained semantic contrastive learning (CL), termed S2CLNet. Departing from the segmentation-driven paradigm, instead of relying solely on segmentation supervision, S2CL introduces region-level structural constraints to perform fine-grained CL in the joint embedding space. This design encourages semantically corresponding visual regions and linguistic representations to be more closely aligned while effectively separating target regions from irrelevant background areas. Furthermore, we develop a dynamic modality balancing module (DMBM) to enhance cross-modal interaction under complex remote sensing scenarios. The DMBM jointly models bidirectional cross-modal attention (BCA) and dynamically adjusts the relative contributions of visual and linguistic modalities according to scene complexity and linguistic specificity, thereby facilitating more adaptive and robust multimodal feature integration. Extensive experiments on three public benchmarks, RefSegRS, RRSIS-D, and RISBench, demonstrate that the proposed method achieves superior performance compared with state-of-the-art approaches. The code will be publicly available at https://github.com/45degreesl
Sen Lei, Shuai Li, Xin-Yu Xiao et al.· IEEE Transactions on Geoscie...· 0 citations
Pixel-level annotation of remote sensing imagery is costly, motivating weakly supervised semantic segmentation (WSSS) using only image-level labels. However, class activation maps (CAMs) often highlight only discriminative sub-regions and fail to separate adjacent land-cover regions, particularly in remote sensing scenes characterized by densely co-occurring land-cover classes and substantial variations in object scale. To address these limitations, we propose a three-stage framework that integrates complementary priors from Contrastive Language–Image Pre-training (CLIP), Self-Distillation with No Labels version 2 (DINOv2), and the Segment Anything Model (SAM). First, a lightweight CLIP adapter aligns vision–language priors with remote sensing imagery, while sigmoid-based multi-label decoupled distillation replaces class-competitive distillation with independent class-wise supervision, producing more complete CAMs. Second, DINOv2-guided feature clustering decomposes large merged regions before SAM prompt generation, while Spatial–Semantic Constraints are used to construct confidence-guided point-and-box prompts and reject excessively expanded or semantically inconsistent masks, thereby generating reliable pseudo-labels. Finally, a compact segmentation network is initialized with the weights learned in Stage 1 and retrained using the refined pseudo-labels generated in Stage 2, eliminating the need for foundation models during inference. Experiments on the Potsdam, LoveDA, and DeepGlobe datasets show that the proposed method achieves mean intersection over union (mIoU) scores of 53.16%, 52.66%, and 62.98%, respectively, outperforming state-of-the-art WSSS baselines by 6.55, 1.16, and 1.27 percentage points, respectively. These results demonstrate the effectiveness and generalizability of the proposed framework across diverse remote sensing scenarios under image-level supervision.
A lightweight and efficient framework that integrates CLIP and DINO foundation models to address three challenges: semantic misalignment between generic text prompts and RSI-specific visuals; static CAM quality; and incomplete object coverage is proposed.
Xin Li, Nicola Genzano, M. Gianinetto et al.· ISPRS Annals of the Photogra...· 0 citations
Remote sensing visual question answering (RSVQA) remains challenging because the evidence relevant to the question in remote sensing (RS) imagery is typically sparse, spatially dispersed, and highly variable in scale and layout. This structural heterogeneity poses a major challenge to existing transfer strategies, which often struggle to achieve effective query-conditioned semantic focusing and localized spatial refinement. To address this issue, we propose CRISP, a task-specific parameter-efficient adaptation framework for RSVQA built on a frozen ViLT backbone. CRISP comprises two complementary components. First, a cross-modal residual guidance (CMRG) module generates instance-specific guidance tokens from pooled image and question summaries, steering early cross-modal interaction toward query-relevant content while suppressing background interference. Second, an attention-guided spatial realignment (ASR) module performs offset-guided feature realignment within intermediate Transformer layers, enabling localized refinement of spatial evidence under scale variation and sparse semantic distribution. Extensive experiments on the RSVQA-LR and RSVQA-HR benchmarks show that CRISP achieves strong overall performance and consistently improves overall accuracy (OA) and average accuracy (AA) over prior methods, with particularly notable gains on presence, comparison, and region-related questions. These results demonstrate that residual guidance and spatial realignment together provide an effective task-specific parameter-efficient adaptation strategy for RSVQA under the frozen-backbone setting. The code will be available at https://github.com/PhD-Xu/CRISP
Chang Xu, Zhongle Ren, Biao Hou et al.· IEEE Transactions on Geoscie...· 0 citations
With the rapid growth of Earth observation technologies, remote sensing archives are rapidly expanding, making remote sensing image-text retrieval (RS-ITR) increasingly important. However, continual RS-ITR remains challenging because scale variation and distribution shifts in RS aggravate cross-modal alignment space distortion, making it difficult for existing continual learning (CL) methods to support reliable continual retrieval. To address this challenge, we propose DARAD, a dual-adapter and ranking-aware distillation framework that preserves the historical cross-modal ranking structure while learning new visual and textual concepts from evolving archives. Specifically, the visual branch introduces a spatial fusion adapter, which integrates coarse regional cues and fine-grained patch cues to accommodate RS scale variation while anchoring visual updates to the pretrained alignment space. The textual branch employs multi-expert semantic routing, which separates shared textual semantics from semantically specialized residuals to absorb newly emerging descriptions while constraining global text embedding drift. Furthermore, bidirectional ranking distillation uses a frozen teacher model and historical anchors to preserve the historical cross-modal ranking structure, thereby mitigating alignment space distortion across continual stages. Experiments under a multi-stage continual retrieval protocol show that DARAD achieves superior performance over existing CL methods, improving adaptation to newly arrived data while maintaining effectiveness on historical data.
Xi Chen, Xu Chen, Xiangyang Jia et al.· 0 citations
State space models, especially Visual State Space Duality (VSSD), have emerged as efficient linear-time alternatives to Transformers for dense visual tasks. However, we observe that VSSD compresses spatial context into a global aggregation that suppresses high-frequency responses, causing excessive boundary smoothing in remote sensing semantic segmentation. To address this, we propose CRISP, a calibration framework with two components. Its core, the Duality Calibration Operator (DCO), restores local contrast and boundary responses through residual injection and frequency calibration within the VSSD backbone, without altering its linear complexity. To retain the recovered detail, an Orthogonal Multi-Prototype (OMP) head assigns multiple orthogonally constrained prototypes per class to model large intra-class variance. Extensive experiments on Potsdam, Vaihingen, and LoveDA show that, with approximately 30M parameters, CRISP achieves consistent gains in mean F1 (mF) and mIoU while remaining competitive with state-of-the-art methods. Code is available at https://github.com/crazylifeha/CRISP.