Remote Sensing Benchmarks

Geo-Foundation Models (GFMs) have been evaluated across diverse Earth observation tasks and domains, showing strong potential to produce reliable maps even with sparse labels. However, systematic benchmarking of GFMs for Cryosphere applications remains limited, primarily because suitable evaluation datasets are scarce. We address this gap by introducing Cryo-Bench, a benchmark comprising six semantic segmentation datasets covering five cryospheric components: supraglacial debris, glacial lakes under two sensing configurations, sea ice, calving fronts and Antarctic ice-shelf extent. The benchmark includes multispectral, RGB, and synthetic aperture radar observations from regions underrepresented in existing pretraining archives. We evaluate thirteen GFMs alongside U-Net and Vision Transformer baselines trained from scratch under a unified evaluation protocol. With frozen encoders, the U-Net achieves the highest six-dataset average mean intersection over union (mIoU) of 68.22%, slightly exceeding TerraMind (67.86%). However, the paired difference of 0.36 percentage points has a 95% confidence interval of [-0.15, +0.89], indicating that the observed ordering is not statistically significant. Fine-tuning with a fixed learning rate produces mixed outcomes, with leading models such as TerraMind and DOFA experiencing declines in average mIoU (-1.2 and -1.7). In contrast, learning-rate optimization substantially improves fine-tuning performance: DOFA reaches 93.97% mIoU on the RGB glacial lake task, ranking the U-Net fourth, while Scale-MAE and GFM-Swin surpass U-Net on calving fronts. In the few-shot setting, five GFMs outperform U-Net, retaining 91.1% of their full-label accuracy compared with 86.1% for U-Net.

Recent research in geospatial machine learning demonstrates that models pretrained with self-supervised learning on Earth observation data can perform well on downstream tasks with limited labeled data. However, most benchmark datasets have few data modalities and poor global representation, limiting the ability to evaluate multimodal pretrained models at global scales. In order to fill this gap, we introduce MMEarth-Bench, a collection of five new environmental tasks with 12 modalities, globally distributed data, and both random and geographic test splits. We benchmark a diverse set of pretrained models and find that while (multimodal) pretraining tends to improve model robustness in limited data settings, geographic generalization abilities remain poor. Moreover, a simple randomly initialized multimodal model is competitive given enough labeled data. Although data is abundant, models can currently only make use of the modalities on which they were pretrained. To solve this problem, we propose using all the modalities available at test time as auxiliary tasks for test-time adaptation. Our model-agnostic method for test-time training with multimodal reconstruction (TTT-MMR) can improve performance across all models and tasks on both test splits. Furthermore, geographic batching leads to a good trade-off between regularization and specialization during TTT, which is especially beneficial for long-tail distributions. Our dataset, code, and visualization tool are linked on the project page: lgordon99.github.io/mmearth-bench.

Vision--language models (VLMs) show promise for disaster-response remote sensing, but existing benchmarks mainly emphasize scene-level or damage-centric assessment. To study this building-centric gap, we introduce \method{}, a diagnostic benchmark built on xBD, a pre/post-disaster satellite dataset with building-level damage labels. \method{} enriches building instances with OpenStreetMap-derived functional labels and contains 134{,}108 task-specific instruction records across 15 task types, spanning instance-level assessment, scene-level counting, multi-instance reasoning, and structured report generation. The benchmark supports RGB pre/post-disaster imagery, single- and multi-view instance formulations, and scene-level RGB/SAR diagnostic inputs. Experiments with general-domain and remote-sensing VLMs show that models perform better on visible damage cues than on building-function understanding, multi-instance reasoning, counting, and grounded reporting. Instruction tuning improves performance on several tasks but does not close this building-centric gap.

Satellite imagery differs from natural images in viewpoint, resolution, scale variation, and the prevalence of small objects -- demanding both region-level spatial reasoning and holistic scene understanding. Existing remote-sensing approaches are fragmented: dual-encoder retrieval models scale well but cannot interleave modalities, whereas generative assistants support grounding, yet are inefficient for retrieval. Benchmarks mirror this split: interleaved evaluations mainly target generative assistants, while cross-modal retrieval benchmarks target dual encoders. To bridge this gap, we introduce \textbf{RSMEB}, a unified remote sensing benchmark that evaluates cross-modal and interleaved retrieval across 21 tasks under a single ranking protocol, enabling comprehensive comparison of retrieval models on region- and geo-aware capabilities as well as conventional retrieval. As a strong reference baseline, we present \textbf{VLM2GeoVec}, an instruction-conditioned, single-encoder interleaving formulation tailored to remote sensing that packs image, text, bounding-box, and geo-coordinate tokens into one sequence and learns a unified embedding via contrastive training. Across RSMEB, VLM2GeoVec achieves $\textbf{26.6\%}$ P@1 in region-caption retrieval ($\textbf{+25}$ percentage points), $\textbf{32.5\%}$ in referring-expression retrieval ($\textbf{+19}$), and $\textbf{17.8\%}$ in semantic geo-aware retrieval (\textbf{>3}$$\times prior best), while remaining competitive in conventional scene classification and text--image retrieval in zero-shot settings. Together, the proposed suite and reference baseline standardize evaluation and deliver a unified embedder for scalable retrieval and region-/geo-aware grounding. The code, the model checkpoints, and the data are available at https://github.com/emasa/VLM2GeoVec.

The data for remote sensing is constantly acquired, and new data comes from a growing number and diversity of satellites, while the vast majority of labeled data comes from older satellites. As remote-sensing foundation models for Earth observation scale up, the cost of (re-)training to support new satellites grows too, so cross-band generalization across sensors and satellites is increasingly important. We introduce GeoCrossBench, an extension of the popular GeoBench benchmark with a new evaluation protocol for cross-band generalization across sensors and satellites: it tests standard in-distribution performance with the same bands for train and test, generalization to inputs with no intersection between train and test; and generalization to test inputs containing a superset of the training bands. We develop $χ$ViT, a self-supervised extension of the band-agnostic ChannelViT, as a supporting baseline for cross-band generalization. We evaluate a representative set of remote-sensing-specific and general-purpose vision models, characterize current performance, and identify directions for improvement through 11,900 H100 GPU-hours of experiments. When averaging dataset-specific metric scores, DOFA leads the in-distribution setting (61.30), frozen Panopticon leads the no-overlap setting (22.75), and ImageNet-pretrained ViT-B leads both the superset setting (56.19) and the overall average across settings (45.27). While top rankings in each setting are close, we clearly see that all models suffer significant performance losses when evaluated on unseen bands. We will publicly release the code and datasets to support the development of more future-proof remote sensing models with stronger cross-band generalization.

We present the P$^3$ dataset, a large-scale multimodal benchmark for building vectorization, constructed from aerial LiDAR point clouds, high-resolution aerial imagery, and vectorized 2D building outlines, collected across three continents. The dataset contains over 10 billion LiDAR points with decimeter-level accuracy and RGB images at a ground sampling distance of 25 centimeter. While many existing datasets primarily focus on the image modality, P$^3$ offers a complementary perspective by also incorporating dense 3D information. We demonstrate that LiDAR point clouds serve as a robust modality for predicting building polygons, both in hybrid and end-to-end learning frameworks. Moreover, fusing aerial LiDAR and imagery further improves accuracy and geometric quality of predicted polygons. The P$^3$ dataset is publicly available, along with code and pretrained weights of three state-of-the-art models for building polygon prediction at https://github.com/raphaelsulzer/PixelsPointsPolygons .