A Composition-Aware Pretraining Framework for Geospatial Foundation Models

• URL: http://arxiv.org/abs/2608.30817v1
• Date: Mon, 31 Aug 2026 14:00:12 GMT
• Title: A Composition-Aware Pretraining Framework for Geospatial Foundation Models
• Abstract summary: We propose a composition-aware pretraining framework that explicitly encodes fractional land-cover mixtures.<n>Each satellite image cell is mapped to a histogram representing its fractional land-cover distribution, which we term the "composition target"<n> Experimental evaluation shows that composition-aware pretraining yields substantial gains on region-level understanding tasks.
• Score: 0.6999740786886535
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Geospatial foundation models have emerged as state-of-the-art methods for downstream Earth observation tasks. However, existing pretraining methodologies process imagery through a single-concept lens, failing to capture the highly compositional nature of complex satellite scenes. We propose a composition-aware pretraining framework that explicitly encodes fractional land-cover mixtures. Each satellite image cell is mapped to a histogram representing its fractional land-cover distribution, which we term the "composition target". These targets serve as the primary prediction objective and are distilled into the backbone using Earth Mover's Distance. Experimental evaluation shows that composition-aware pretraining yields substantial gains on region-level understanding tasks requiring semantic similarity judgment, including zero-shot image retrieval and scene classification, while remaining competitive on tasks requiring fine-grained spatial precision, such as segmentation and object detection. With a 36.8M-parameter backbone, our framework outperforms SatMAE and Prithvi-EO-2.0, which contain 303M and 600M parameters, respectively, in most retrieval and scene classification settings. On the fine-grained ForestNet-12 dataset, a rigorous testbed for compositional discrimination, our method boosts baseline mAP@10 from 0.279 to 0.434, a 55.6% relative improvement, providing direct evidence for the effectiveness of explicit composition modeling. The code implementation can be found at https://github.com/05kashyap/GFM_Composition_Pretraining

Related papers

• Back to the Feature: Zero-Shot 6DoF Pose Estimation via Dense Local Features [0.0]
  We present B2ose, a training-free zero-shot method for 6DoF pose estimation of unseen objects from RGB images.<n>B2ose extracts dense patch-level features that generalize across the synthetic-to-real domain gap without task-specific fine-tuning.<n>On the seven core datasets of the BOP Benchmark, B2ose achieves 40.7 mean AR without refinement and 56.4 refinement.
  arXiv  Detail & Related papers  (2026-09-06T16:57:27Z)
• Ground-to-Satellite Localization in Unconstrained Image Collections for 3D Scene Reconstruction [9.370818606626663]
  Ground image localization with respect to satellite imagery is a key enabler for metrically-accurate, geo-localized 3D scene reconstruction.<n>Existing cross-view localization methods have strict requirements such as panoramic imagery or known initial locations.<n>We propose a robust hierarchical cross-view localization framework that leverages geometric constraints from Structure-from-Motion (SfM) models.
  arXiv  Detail & Related papers  (2026-08-29T11:52:21Z)
• AerialMetric: Benchmarking and Adapting UAV Monocular Metric Depth Estimation in the Real World [75.54693105865003]
  We introduce AerialMetric, a benchmark dataset to evaluate and facilitate the adaptation of monocular metric depth estimation under UAV aerial viewpoints.<n>AerialMetric provides 52K real-world and 16K synthetic image-depth pairs with reliable metric ground truth.
  arXiv  Detail & Related papers  (2026-06-29T02:48:47Z)
• Component-Aware Structure-Preserving Style Transfer for Satellite Visual Sim2Real Data Construction [7.240627719297504]
  This paper presents a structure-preserving style transfer framework for satellite visual synthetic-to-real data construction.<n>The proposed method achieves the lowest image distribution discrepancy, with an FID of 54.32 and a KID of 0.048.
  arXiv  Detail & Related papers  (2026-05-19T10:01:17Z)
• Loc$^2$: Interpretable Cross-View Localization via Depth-Lifted Local Feature Matching [80.57282092735991]
  We propose an accurate and interpretable fine-grained cross-view localization method.<n>It estimates the 3 Degrees of Freedom (DoF) pose of a ground-level image by matching its local features with a reference aerial image.<n> Experiments show state-of-the-art accuracy in challenging scenarios such as cross-area testing and unknown orientation.
  arXiv  Detail & Related papers  (2025-09-11T18:52:16Z)
• Zero-shot Inexact CAD Model Alignment from a Single Image [53.37898107159792]
  A practical approach to infer 3D scene structure from a single image is to retrieve a closely matching 3D model from a database and align it with the object in the image.<n>Existing methods rely on supervised training with images and pose annotations, which limits them to a narrow set of object categories.<n>We propose a weakly supervised 9-DoF alignment method for inexact 3D models that requires no pose annotations and generalizes to unseen categories.
  arXiv  Detail & Related papers  (2025-07-04T04:46:59Z)
• Position Prediction Self-Supervised Learning for Multimodal Satellite Imagery Semantic Segmentation [0.0]
  We propose adapting LOCA (Location-aware), a position prediction self-supervised learning method, for multimodal satellite imagery semantic segmentation.<n>Our approach addresses the unique challenges of satellite data by extending SatMAE's channel grouping from multispectral to multimodal data.<n>We evaluate our approach on the Sen1Floods11 flood mapping dataset, where it significantly outperforms existing reconstruction-based self-supervised learning methods for satellite imagery.
  arXiv  Detail & Related papers  (2025-06-07T16:16:29Z)
• Classifying geospatial objects from multiview aerial imagery using semantic meshes [2.116528763953217]
  We propose a new method to predict tree species based on aerial images of forests in the U.S. We show that our proposed multiview method improves classification accuracy from 53% to 75% relative to an orthoorthoaic baseline on a challenging cross-site tree classification task.
  arXiv  Detail & Related papers  (2024-05-15T17:56:49Z)
• Beyond Cross-view Image Retrieval: Highly Accurate Vehicle Localization Using Satellite Image [91.29546868637911]
  This paper addresses the problem of vehicle-mounted camera localization by matching a ground-level image with an overhead-view satellite map. The key idea is to formulate the task as pose estimation and solve it by neural-net based optimization. Experiments on standard autonomous vehicle localization datasets have confirmed the superiority of the proposed method.
  arXiv  Detail & Related papers  (2022-04-10T19:16:58Z)
• Mars Terrain Segmentation with Less Labels [1.1745324895296465]
  This research proposes a semi-supervised learning framework for Mars terrain segmentation. It incorporates a backbone module which is trained using a contrastive loss function and an output atrous convolution module. The proposed model is able to achieve a segmentation accuracy of 91.1% using only 161 training images.
  arXiv  Detail & Related papers  (2022-02-01T22:25:15Z)
• Context-self contrastive pretraining for crop type semantic segmentation [39.81074867563505]
  The proposed Context-Self Contrastive Loss (CSCL) learns an embedding space that makes semantic boundaries pop-up. For crop type semantic segmentation from Satellite Image Time Series (SITS) we find performance at parcel boundaries to be a critical bottleneck. We present a process for semantic segmentation at super-resolution for obtaining crop classes at a more granular level.
  arXiv  Detail & Related papers  (2021-04-09T11:29:44Z)
• A New Mask R-CNN Based Method for Improved Landslide Detection [54.7905160534631]
  This paper presents a novel method of landslide detection by exploiting the Mask R-CNN capability of identifying an object layout. A data set of 160 elements is created containing landslide and non-landslide images. The proposed algorithm can be potentially useful for land use planners and policy makers of hilly areas.
  arXiv  Detail & Related papers  (2020-10-04T07:46:37Z)
• Grounded Situation Recognition [56.18102368133022]
  We introduce Grounded Situation Recognition (GSR), a task that requires producing structured semantic summaries of images. GSR presents important technical challenges: identifying semantic saliency, categorizing and localizing a large and diverse set of entities. We show initial findings on three exciting future directions enabled by our models: conditional querying, visual chaining, and grounded semantic aware image retrieval.
  arXiv  Detail & Related papers  (2020-03-26T17:57:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.