GeoReason: Bridging Logical Reasoning and Spatial Fidelity in Remote Sensing Segmentation

Cheng Fan; Han Wu; Tingzhang Luo; Minjing Dong; Yehui Tang; Jianyuan Guo

cs.CVOct 16, 2026

GeoReason: Bridging Logical Reasoning and Spatial Fidelity in Remote Sensing Segmentation

Authors: Cheng Fan, Han Wu, Tingzhang Luo, Minjing Dong, Yehui Tang, Jianyuan Guo

Organizations: City University of Hong Kong · School of Computer Science, Peking University · Samsung

Abstract

Multimodal Large Language Models (MLLMs) have significantly advanced instruction-driven perception through the "Embedding-as-Mask" paradigm. However, extending this capability to Remote Sensing (RS) reasoning segmentation remains exceptionally challenging under cluttered geographical contexts and extreme scale variations, often leading to Spatial Fidelity Degradation and Target Misidentification. We identify a critical yet underexplored bottleneck: the sequential alienation of the segmentation token [SEG] from its initial visual grounding within the autoregressive generation paradigm. Produced only at the end of long reasoning chains, [SEG] relies on visual evidence that has been progressively diluted by linguistic abstraction and autoregressive noise. This information decay is particularly catastrophic for RS targets with minuscule pixel footprints, where even marginal fidelity loss leads to localization failure. To address these bottlenecks, we propose GeoReason, a novel framework designed to fortify logical deduction and preserve spatial integrity. First, we introduce an Anticipatory Prior mechanism that injects the [SEG] token directly into the initial user query, shifting it from a terminal output to a primary condition. This paradigm shift transforms the token into a persistent spatial anchor, preventing information decay during long-form linguistic inference. Second, we enhance visual reasoning via a Location-aware Chain-of-Thought (CoT) that enforces coarse-to-fine localization using bounding boxes, coupled with a Latent Patch-to-Token Alignment (PTA) Loss that explicitly aligns the latent segmentation token with physical RS textures. Extensive experiments demonstrate that GeoReason consistently outperforms state-of-the-art methods on the Earthreason and RRSIS-D benchmarks. Code is available at https://github.com/Fanc0126/GeoReason.

Figures & tables

Figure 1: Qualitative comparison of GeoReason against the baseline. (Top) While the baseline identifies the viaduct, the resulting mask is fragmented and suffers from blurry boundaries. (Bottom) The baseline fails to interpret the functional implications of the query (“sports games or popular live music events”), leading to the segmentation of an irrelevant field.

Figure 2: Spatial fidelity degradation of masks induced by long reasoning chains.

Figure 3: Overview of the GeoReason framework. (Left) Overall architecture of Anticipatory Prior mechanism, where a [SEG] token is injected into the user query to pre-anchor spatial evidence and mitigate information decay during long-chain reasoning. The MLLM is trained to output intermediate bounding boxes as spatial anchors ( Location-aware CoT ) before generating the final [SEG] . (Right) The Latent Patch-to-Token Alignment (PTA) loss aligns the hidden states of both prompt-side and answer-side [SEG] tokens with the visual patch embeddings corresponding to the GT target.

Figure 4: Illustration of the prompt template and model response format.

Table 5

Panel of Figure 5 — Figure 5: Token-level mechanism analysis. (a) Direct attention from different [SEG] token sources to image tokens. (b) Foreground/background alignment and token-only mask quality.

Panel of Figure 6 — Figure 6: Case-level evidence. (a) GeoReason preserves focus on the target region and produces a cleaner mask than the baseline. (b) In the same case, the final answer-side [SEG] focuses strongly on the prompt-side [SEG] .

Components			Earthreason (Reasoning)				RRSIS-D (Referring)
AP	BBox	PTA	Val g	Val c	Test g	Test c	Val g	Val c	Test g	Test c
✗	✗	✗	75.04	70.33	75.32	70.45	67.13	79.39	66.42	78.67
✗	✗	✓	76.02	72.26	76.32	73.01	67.98	79.71	66.99	79.32
✗	✓	✗	74.42	70.89	74.54	70.59	65.60	76.13	65.29	76.43
✓	✗	✗	76.24	77.32	78.02	78.87	67.54	79.45	66.13	78.32
✓	✗	✓	76.45	75.64	77.39	77.18	68.12	79.88	66.32	78.67

Table 3: Ablation study of GeoReason components. AP: Anticipatory Prior, BBox: Location-aware CoT, PTA: Latent PTA Loss. g: gIoU, c: cIoU (%).

Table 9

Figure 7: Reasoning-sensitive case study on EarthReason. GeoReason correctly selects the protected storage structure required for safe raw-material storage, whereas the Zero-Reasoning variant overfits to a visually plausible but functionally incorrect open yard.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Prefix (%)	Image-attention mass		Hidden-state drift
Prefix (%)	Baseline	GeoReason	Baseline	GeoReason
0	0.2389	0.2593	0.0000	0.0000
25	0.2206	0.2502	0.1955	0.0233
50	0.2135	0.2491	0.1896	0.0248
75	0.2199	0.2481	0.2097	0.0280
100	0.2003	0.2446	0.2076	0.0495

Appendix

Table 6: Effect of reasoning-prefix length on the answer-side [SEG] on EarthReason test. Image-attention mass is summed over image tokens and averaged across heads, the last four layers, and instances. Hidden-state drift is the mean cosine distance between the last-layer [SEG] representations at the retained and zero-length prefixes.

Method / Variant	RRSIS-D (Referring)				EarthReason (Reasoning)
Method / Variant	Val gIoU	Val cIoU	Test gIoU	Test cIoU	Val gIoU	Val cIoU	Test gIoU	Test cIoU
GeoReason	68.36	79.87	67.32	79.39	77.61	77.63	78.61	80.60
w/o PTA Head	68.26 -0.10	78.26 -1.61	67.24 -0.08	78.38 -1.01	77.20 -0.41	75.92 -1.71	77.31 -1.30	75.93 -4.67
Simple PTA Head (MLP)	68.39 +0.03	78.85 -1.02	67.46 +0.14	78.80 -0.59	77.51 -0.10	78.60 +0.97	78.34 -0.27	79.78 -0.82
BBox-based Alignment	68.56 +0.20	79.26 -0.61	67.28 -0.04	78.65 -0.74	77.26 -0.35	77.71 +0.08	78.64 +0.03	79.47 -1.13
PTA Pooling: Attention	67.76 -0.60	79.05 -0.82	66.69 -0.63	78.11 -1.28	77.47 -0.14	77.47 -0.16	78.10 -0.51	78.65 -1.95

Appendix

Table 7: Ablation studies on PTA Loss configurations.

\Delta

denotes the performance gap relative to our final GeoReason model (red for decrease, blue for increase). Results are reported in (%).

Method	FT	Referring		Reasoning
Method	FT	gIoU	cIoU	gIoU	cIoU
LISA [ 15 ]	✗	4.56	3.53	5.77	7.09
PixelLM [ 37 ]	✗	5.80	6.04	6.30	6.36
PSALM [ 66 ]	✗	18.91	31.77	9.27	11.88
HIPIE [ 46 ]	✗	39.14	29.25	11.76	7.90
SegLLM [ 47 ]	✗	35.27	15.90	16.41	11.06

Appendix

Table 8: Results on GeoSeg-Bench. FT indicates fine-tuning on the GeoSeg-1M training dataset [ 31 ] . Baseline scores are reproduced from UniGeoSeg [ 31 ] . Best results are in bold ; second-best are underlined .

Figure 8: Qualitative case study comparing the baseline and GeoReason. GeoReason first grounds the correct region with an intermediate bbox in CoT, then exhibits stronger target-focused [SEG] and produces a more accurate final mask.

Reasoning Segmentation

Reasoning segmentation links language and visual reasoning to pixel-level masks for targets defined by complex intent, relations, or context.

See all 33 papers

Mar 4, 2026

GeoSeg: Training-Free Reasoning-Driven Segmentation in Remote Sensing Imagery

Recent advances in MLLMs are reframing segmentation from fixed-category prediction to instruction-grounded localization. While reasoning based segmentation has progressed rapidly in natural scenes, remote sensing lacks a generalizable solution due to the prohibitive cost of reasoning-oriented data and domain-specific challenges like overhead viewpoints. We present GeoSeg, a zero-shot, training-free framework that bypasses the supervision bottleneck for reasoning-driven remote sensing segmentation. GeoSeg couples MLLM reasoning with precise localization via: (i) bias-aware coordinate refinement to correct systematic grounding shifts and (ii) a dual-route prompting mechanism to fuse semantic intent with fine-grained spatial cues. We also introduce GeoSeg-Bench, a diagnostic benchmark of 810 image--query pairs with hierarchical difficulty levels. Experiments show that GeoSeg consistently outperforms all baselines, with extensive ablations confirming the effectiveness and necessity of each component.

Lifan Jiang, Yuhang Pei, oxi Wu +5

State Key lab of CAD&CG, Zhejiang University · UniTTEC Co. Ltd. · Wuchan Zhongda Chengtou (Ningbo) Holdings. Ltd.

Remote Sensing Image Segmentation Reasoning Segmentation Semantic Segmentation Remote Sensing Benchmarks

Ask PDF

Aug 14, 2026

FIRM: Fine-Grained Intra-Token Representation of Masks for Remote Sensing Reasoning Segmentation

Reasoning segmentation requires multimodal large language models (MLLMs) to translate implicit instructions into precise pixel-level masks. MLLMs encode an image as visual tokens, each of which merges a group of image patches. In remote sensing images, small targets, thin structures, and adjacent instances can occupy different parts of the same visual token. Assigning a single binary mask label to such a token loses its internal spatial structure, causing nearby targets to merge and object boundaries to become coarse. To bridge this representational gap, we introduce FIRM, a Fine-grained Intra-token Representation of Masks. For each visual token, FIRM predicts a mask code that specifies an

r\times r

binary sub-cell pattern rather than a single foreground/background label. Given a target identified by the MLLM, the complete grid of mask codes is predicted in one mask pass. Fixed lookup converts the predicted codes into a discrete sub-cell mask, while marginalizing the code distribution yields a soft structural field. To further recover fine-grained boundaries within each sub-cell, we introduce a lightweight continuous renderer that refines this field using pre-merge visual features and image details. Across five reasoning and referring segmentation benchmarks on satellite and UAV images, FIRM achieves leading results, including

70.5/80.5

gIoU/cIoU on LaSeRS and a

3.0

-point average gain on EarthReason. These results demonstrate the value of explicitly representing intra-token mask patterns for fine-grained MLLM segmentation.

Weidong Tang, Kaiyu Li, Yikai Wang +4

Xi’an Jiaotong University, Xi’an, China · China Agricultural University, Beijing, China · Renmin University of China, Beijing, China +1

Remote Sensing Image Segmentation Referring Image Segmentation Boundary-Aware Segmentation

Ask PDF

Jul 6, 2026

DGSeg: Dynamic Gating of Semantic-Spatial Guided Predictions for Reasoning Segmentation

Reasoning segmentation aims to predict pixel-wise masks for targets given complex language queries. Existing approaches leverage Multimodal Large Language Models (MLLMs) for vision-language reasoning and generate intermediate target cues (e.g., points or boxes) to guide a segmentation model. However, compressing rich reasoning into sparse cues often introduces ambiguity and noise, preventing these cues from accurately preserving the reasoning intent. While multiple complementary cues can enrich target information, existing methods typically feed them jointly into a single segmentation process, allowing ambiguous or erroneous cues to affect the entire prediction. Therefore, we propose DGSeg, a reasoning segmentation framework that learns to fuse predictions guided by semantic and spatial cues. Specifically, the MLLM jointly reasons about both target identity and spatial location, producing complementary semantic and spatial cues that are fed into separate segmentation branches. Their predictions are adaptively integrated by a lightweight dynamic gating module trained with relative branch-quality supervision to suppress noisy or conflicting regions. Extensive experiments demonstrate that DGSeg consistently outperforms strong baselines on multiple benchmarks and achieves 69.6% and 67.3% gIoU on the challenging ReasonSeg validation and test splits. Code is available at https://github.com/RZZeng/DGSeg.

Ruizhe Zeng, Siyu Cao, Lu Zhang +1

State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China · School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China · Nanjing Artificial Intelligence Research of IA, Nanjing, 211100, China

Text-Guided Image Segmentation VLM Reasoning Reasoning Segmentation

Ask PDF

Abstract

Reasoning Segmentation

Explore similar work

GeoSeg: Training-Free Reasoning-Driven Segmentation in Remote Sensing Imagery

FIRM: Fine-Grained Intra-Token Representation of Masks for Remote Sensing Reasoning Segmentation

DGSeg: Dynamic Gating of Semantic-Spatial Guided Predictions for Reasoning Segmentation