arXiv subjects

Zexian Huang

Publications and source records attributed to Zexian Huang.

4 recordsLinked to original sources

arXiv2026

LMM Modality Transfer: A Pre-requisite for Autonomous GIS Agents

AI models are becoming increasingly adept at understanding and processing spatial information, thereby facilitating agentic problem-solving in spatial tasks and workflows. However, most of the research on their spatial capabilities (e.g., spatial reasoning) has focused on the textual modality as input and output. This contrasts with the human approach to GIS workflows, where text and visual modalities are often used together, interchangeably, and in a complementary manner. Thus, to truly achieve an automated GIS analysis pipeline or carry out human-designed GIS workflows, AI models --- Large Multimodal Models (LMMs) in particular --- need to be able to seamlessly transition between image- and text-based modalities that are traditionally used in such workflows. We present a modality transfer task that (1) asks an LMM to first describe an input image of colored squares in a regular grid, and (2) asks a new LMM instance to re-generate an image of the original spatial scene using the textual description output by the former model. This task quantifies the ability of LMMs to transfer spatial information between image and text modalities. Ultimately, by examining the modality transfer capability of LMMs through the lens of spatial information theory, this work highlights a critical bottleneck: achieving strong and robust geospatial understanding in LMMs requires rigorous, multi-modal alignment. Our results indicate that recent LMMs (here from OpenAI) still struggle with modality transfer, when tasked with re-generating an image of a simple spatial grid of color squares.

cs.AI↗

arXiv2026

Mapping Hidden Heritage: Self-supervised Pre-training on High-Resolution LiDAR DEM Derivatives for Archaeological Stone Wall Detection

Historic dry-stone walls hold significant cultural and environmental importance, serving as historical markers and contributing to ecosystem preservation and wildfire management during dry seasons in Australia. However, many of these stone structures in remote or vegetated landscapes remain undocumented due to limited accessibility and the high cost of manual mapping. Deep learning-based segmentation offers a scalable approach for automated mapping of such features, but challenges remain: 1.the visual occlusion of low-lying dry-stone walls by dense vegetation and 2.the scarcity of labeled training data. This study presents DINO-CV, a self-supervised cross-view pre-training framework based on knowledge distillation, designed for accurate and data-efficient mapping of dry-stone walls using Digital Elevation Models (DEMs) derived from high-resolution airborne LiDAR. By learning invariant geometric and geomorphic features across DEM-derived views, (i.e., Multi-directional Hillshade and Visualization for Archaeological Topography), DINO-CV addresses the occlusion by vegetation and data scarcity challenges. Applied to the Budj Bim Cultural Landscape at Victoria, Australia, a UNESCO World Heritage site, the approach achieves a mean Intersection over Union (mIoU) of 68.6% on test areas and maintains 63.8% mIoU when fine-tuned with only 10% labeled data. These results demonstrate the potential of self-supervised learning on high-resolution DEM derivatives for large-scale, automated mapping of cultural heritage features in complex and vegetated environments. Beyond archaeology, this approach offers a scalable solution for environmental monitoring and heritage preservation across inaccessible or environmentally sensitive regions.

cs.CV↗

arXiv2025

LMSeg: An end-to-end geometric message-passing network on barycentric dual graphs for large-scale landscape mesh segmentation

Semantic segmentation of large-scale 3D landscape meshes is critical for geospatial analysis in complex environments, yet existing approaches face persistent challenges of scalability, end-to-end trainability, and accurate segmentation of small and irregular objects. To address these issues, we introduce the BudjBim Wall (BBW) dataset, a large-scale annotated mesh dataset derived from high-resolution LiDAR scans of the UNESCO World Heritage-listed Budj Bim cultural landscape in Victoria, Australia. The BBW dataset captures historic dry-stone wall structures that are difficult to detect under vegetation occlusion, supporting research in underrepresented cultural heritage contexts. Building on this dataset, we propose LMSeg, a deep graph message-passing network for semantic segmentation of large-scale meshes. LMSeg employs a barycentric dual graph representation of mesh faces and introduces the Geometry Aggregation+ (GA+) module, a learnable softmax-based operator that adaptively combines neighborhood features and captures high-frequency geometric variations. A hierarchical-local dual pooling integrates hierarchical and local geometric aggregation to balance global context with fine-detail preservation. Experiments on three large-scale benchmarks (SUM, H3D, and BBW) show that LMSeg achieves 75.1% mIoU on SUM, 78.4% O.A. on H3D, and 62.4% mIoU on BBW, using only 2.4M lightweight parameters. In particular, LMSeg demonstrates accurate segmentation across both urban and natural scenes-capturing small-object classes such as vehicles and high vegetation in complex city environments, while also reliably detecting dry-stone walls in dense, occluded rural landscapes. Together, the BBW dataset and LMSeg provide a practical and extensible method for advancing 3D mesh segmentation in cultural heritage, environmental monitoring, and urban applications.

cs.CV↗

arXiv2025

Learning Geometric Invariant Features for Classification of Vector Polygons with Graph Message-passing Neural Network

Geometric shape classification of vector polygons remains a challenging task in spatial analysis. Previous studies have primarily focused on deep learning approaches for rasterized vector polygons, while the study of discrete polygon representations and corresponding learning methods remains underexplored. In this study, we investigate a graph-based representation of vector polygons and propose a simple graph message-passing framework, PolyMP, along with its densely self-connected variant, PolyMP-DSC, to learn more expressive and robust latent representations of polygons. This framework hierarchically captures self-looped graph information and learns geometric-invariant features for polygon shape classification. Through extensive experiments, we demonstrate that combining a permutation-invariant graph message-passing neural network with a densely self-connected mechanism achieves robust performance on benchmark datasets, including synthetic glyphs and real-world building footprints, outperforming several baseline methods. Our findings indicate that PolyMP and PolyMP-DSC effectively capture expressive geometric features that remain invariant under common transformations, such as translation, rotation, scaling, and shearing, while also being robust to trivial vertex removals. Furthermore, we highlight the strong generalization ability of the proposed approach, enabling the transfer of learned geometric features from synthetic glyph polygons to real-world building footprints.

cs.CV↗