Now We Know? A Systematic Comparison of TerraMind and THOR

• URL: http://arxiv.org/abs/2607.18504v2
• Date: Fri, 24 Jul 2026 09:06:53 GMT
• Title: Now We Know? A Systematic Comparison of TerraMind and THOR
• Abstract summary: This study compares two GFMs developed under European Space Agency's $$-lab with contrasting design philosophies.<n>We find that architectural design choices - patch size and decoder type in particular - explain more performance variance than model identity itself.
• Score: 3.854501808287241
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Benchmarks for Geospatial Foundation Models (GFMs) increasingly rank models by aggregate score, but such rankings obscure why models differ: how much of the gap is architecture, how much is decoder capacity, and how much is a use-case-specific artefact? This study addresses that gap through a controlled comparison of two GFMs developed under European Space Agency's $Φ$-lab with contrasting design philosophies: THOR, which introduces a compute-adaptive architecture supporting variable patch sizes and unifies Sentinel-1, -2, and -3 data at their native resolutions; and TerraMind, a multimodal generative GFM pretrained with a dual-scale token/pixel objective that enables any-to-any cross-modal generation (Thinking-in-Modalities) to infer missing sensors at inference time. Rather than reporting a single leaderboard, we investigate the axes along which the two architectures actually differ - patch size, decoder complexity, finetuning regime, input modality, and model scale - across ten use cases spanning segmentation and regression in diverse domains, including climate disaster response, methane leak detection, snow monitoring, or sea ice mapping. We find that architectural design choices - patch size and decoder type in particular - explain more performance variance than model identity itself, that the two models embody complementary investment strategies (pretraining-time scale for TerraMind versus inference-time tokenisation for THOR), and that correctly interpreting results requires dataset-level characterisation. The resulting picture is not a single winner but a set of hypotheses and a diagnostic ablation methodology that we expect to generalise to future GFMs beyond THOR and TerraMind.

Related papers

• FLORO: A Multimodal Geospatial Foundation Model for Ecological Remote Sensing Across Sensors and Scales [65.4821703903285]
  FLORO is a multimodal foundation model designed to learn transferable representations from a small but highly diverse remote sensing corpus.<n> FLORO is pretrained using masked autoencoding on a heterogeneous combination of Sentinel-1, Sentinel-2, SkySAT imagery, elevation, and UAV-derived data.<n>We evaluated FLORO on the PANGAEA benchmark under a frozen-encoder protocol across scene classification, segmentation, and regression tasks.
  arXiv  Detail & Related papers  (2026-05-27T08:55:54Z)
• THOR: A Versatile Foundation Model for Earth Observation Climate and Society Applications [9.852915112122567]
  THOR is a "computeadaptive" foundation model that solves both input heterogeneity and deployment rigidity.<n>We pre-train THOR with a novel randomized patch and input image size strategy.<n>This allows a single set of pre-trained weights to be deployed at inference with any patch size, enabling a dynamic trade-off between computational cost and feature resolution without retraining.
  arXiv  Detail & Related papers  (2026-01-22T14:38:00Z)
• Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs [63.82840470917859]
  We show that the decoding mechanism of dLLMs can be used as a powerful tool for model attribution.<n>We propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors.
  arXiv  Detail & Related papers  (2025-10-02T06:25:10Z)
• Two Facets of the Same Optimization Coin: Model Degradation and Representation Collapse in Graph Foundation Models [41.39154912912367]
  GFMs are designed to learn the optimal embedding functions from multi-domain text-attributed graphs.<n> graph VQ-MAE stands out among the increasingly diverse landscape of GFM.<n>Despite its potential, domain generalization conflicts cause imperceptible pitfalls.
  arXiv  Detail & Related papers  (2025-09-10T08:45:08Z)
• Intrinsic Explainability of Multimodal Learning for Crop Yield Prediction [36.766406330345525]
  We leverage the intrinsic explainability of Transformer-based models to explain multimodal learning networks.<n>This study focuses on the task of crop yield prediction at the subfield level.
  arXiv  Detail & Related papers  (2025-08-09T11:09:10Z)
• TITAN: Query-Token based Domain Adaptive Adversarial Learning [0.0]
  We focus on the source-free domain adaptive object detection (SF-DAOD) problem when source data is unavailable during adaptation and the model must adapt to an unlabeled target domain.<n>The majority of approaches for the problem employ a self-supervised approach using a student-teacher (ST) framework where pseudo-labels are generated via a source-pretrained model for further fine-tuning.<n>We observe that the performance of a student model often degrades drastically, due to the collapse of the teacher model, primarily caused by high noise in pseudo-labels.<n>To obtain reliable pseudo-labels, we propose
  arXiv  Detail & Related papers  (2025-06-26T17:12:58Z)
• Magnifier: A Multi-grained Neural Network-based Architecture for Burned Area Delineation [4.833815605196964]
  In crisis management and remote sensing, image segmentation plays a crucial role, enabling tasks like disaster response and emergency planning.<n>The problem in their development is the data scarcity and the lack of extensive benchmark datasets, limiting the capabilities of training large neural network models.<n>We propose a novel methodology, namely Magnifier, to improve segmentation performance with limited data availability.
  arXiv  Detail & Related papers  (2025-04-28T08:51:54Z)
• Manifold Topology Divergence: a Framework for Comparing Data Manifolds [109.0784952256104]
  We develop a framework for comparing data manifold, aimed at the evaluation of deep generative models. Based on the Cross-Barcode, we introduce the Manifold Topology Divergence score (MTop-Divergence) We demonstrate that the MTop-Divergence accurately detects various degrees of mode-dropping, intra-mode collapse, mode invention, and image disturbance.
  arXiv  Detail & Related papers  (2021-06-08T00:30:43Z)
• Graphical Modeling for Multi-Source Domain Adaptation [56.05348879528149]
  Multi-Source Domain Adaptation (MSDA) focuses on transferring the knowledge from multiple source domains to the target domain. We propose two types of graphical models,i.e. Conditional Random Field for MSDA (CRF-MSDA) and Markov Random Field for MSDA (MRF-MSDA) We evaluate these two models on four standard benchmark data sets of MSDA with distinct domain shift and data complexity.
  arXiv  Detail & Related papers  (2021-04-27T09:04:22Z)
• Semantic Change Detection with Asymmetric Siamese Networks [71.28665116793138]
  Given two aerial images, semantic change detection aims to locate the land-cover variations and identify their change types with pixel-wise boundaries. This problem is vital in many earth vision related tasks, such as precise urban planning and natural resource management. We present an asymmetric siamese network (ASN) to locate and identify semantic changes through feature pairs obtained from modules of widely different structures.
  arXiv  Detail & Related papers  (2020-10-12T13:26:30Z)
• NAS-Count: Counting-by-Density with Neural Architecture Search [74.92941571724525]
  We automate the design of counting models with Neural Architecture Search (NAS) We introduce an end-to-end searched encoder-decoder architecture, Automatic Multi-Scale Network (AMSNet)
  arXiv  Detail & Related papers  (2020-02-29T09:18:17Z)

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.