Zhitong Xiong

Yang Zhan, Zhitong Xiong, Yuan Yuan · mit

In this paper, we introduce the task of visual grounding for remote sensing data (RSVG). RSVG aims to localize the referred objects in remote sensing (RS) images with the guidance of natural language. To retrieve rich information from RS imagery using natural language, many research tasks, like RS image visual question answering, RS image captioning, and RS image-text retrieval have been investigated a lot. However, the object-level visual grounding on RS images is still under-explored. Thus, in this work, we propose to construct the dataset and explore deep learning models for the RSVG task. Specifically, our contributions can be summarized as follows. 1) We build the new large-scale benchmark dataset of RSVG, termed RSVGD, to fully advance the research of RSVG. This new dataset includes image/expression/box triplets for training and evaluating visual grounding models. 2) We benchmark extensive state-of-the-art (SOTA) natural image visual grounding methods on the constructed RSVGD dataset, and some insightful analyses are provided based on the results. 3) A novel transformer-based Multi-Level Cross-Modal feature learning (MLCM) module is proposed. Remotely-sensed images are usually with large scale variations and cluttered backgrounds. To deal with the scale-variation problem, the MLCM module takes advantage of multi-scale visual features and multi-granularity textual embeddings to learn more discriminative representations. To cope with the cluttered background problem, MLCM adaptively filters irrelevant noise and enhances salient features. In this way, our proposed model can incorporate more effective multi-level and multi-modal features to boost performance. Furthermore, this work also provides useful insights for developing better RSVG models. The dataset and code will be publicly available at https://github.com/ZhanYang-nwpu/RSVG-pytorch.

Fahong Zhang, Yilei Shi, Zhitong Xiong et al.

Object detection is an essential and fundamental task in computer vision and satellite image processing. Existing deep learning methods have achieved impressive performance thanks to the availability of large-scale annotated datasets. Yet, in real-world applications the availability of labels is limited. In this context, few-shot object detection (FSOD) has emerged as a promising direction, which aims at enabling the model to detect novel objects with only few of them annotated. However, many existing FSOD algorithms overlook a critical issue: when an input image contains multiple novel objects and only a subset of them are annotated, the unlabeled objects will be considered as background during training. This can cause confusions and severely impact the model's ability to recall novel objects. To address this issue, we propose a self-training-based FSOD (ST-FSOD) approach, which incorporates the self-training mechanism into the few-shot fine-tuning process. ST-FSOD aims to enable the discovery of novel objects that are not annotated, and take them into account during training. On the one hand, we devise a two-branch region proposal networks (RPN) to separate the proposal extraction of base and novel objects, On another hand, we incorporate the student-teacher mechanism into RPN and the region of interest (RoI) head to include those highly confident yet unlabeled targets as pseudo labels. Experimental results demonstrate that our proposed method outperforms the state-of-the-art in various FSOD settings by a large margin. The codes will be publicly available at https://github.com/zhu-xlab/ST-FSOD.

Yi Wang, Nassim Ait Ali Braham, Zhitong Xiong et al.

Self-supervised pre-training bears potential to generate expressive representations without human annotation. Most pre-training in Earth observation (EO) are based on ImageNet or medium-size, labeled remote sensing (RS) datasets. We share an unlabeled RS dataset SSL4EO-S12 (Self-Supervised Learning for Earth Observation - Sentinel-1/2) to assemble a large-scale, global, multimodal, and multi-seasonal corpus of satellite imagery from the ESA Sentinel-1 \& -2 satellite missions. For EO applications we demonstrate SSL4EO-S12 to succeed in self-supervised pre-training for a set of methods: MoCo-v2, DINO, MAE, and data2vec. Resulting models yield downstream performance close to, or surpassing accuracy measures of supervised learning. In addition, pre-training on SSL4EO-S12 excels compared to existing datasets. We make openly available the dataset, related source code, and pre-trained models at https://github.com/zhu-xlab/SSL4EO-S12.

Zhaiyu Chen, Yilei Shi, Liangliang Nan et al.

We present PolyGNN, a polyhedron-based graph neural network for 3D building reconstruction from point clouds. PolyGNN learns to assemble primitives obtained by polyhedral decomposition via graph node classification, achieving a watertight and compact reconstruction. To effectively represent arbitrary-shaped polyhedra in the neural network, we propose a skeleton-based sampling strategy to generate polyhedron-wise queries. These queries are then incorporated with inter-polyhedron adjacency to enhance the classification. PolyGNN is end-to-end optimizable and is designed to accommodate variable-size input points, polyhedra, and queries with an index-driven batching technique. To address the abstraction gap between existing city-building models and the underlying instances, and provide a fair evaluation of the proposed method, we develop our method on a large-scale synthetic dataset with well-defined ground truths of polyhedral labels. We further conduct a transferability analysis across cities and on real-world point clouds. Both qualitative and quantitative results demonstrate the effectiveness of our method, particularly its efficiency for large-scale reconstructions. The source code and data are available at https://github.com/chenzhaiyu/polygnn.

Yi Wang, Conrad M Albrecht, Nassim Ait Ali Braham et al.

The increasing availability of multi-sensor data sparks wide interest in multimodal self-supervised learning. However, most existing approaches learn only common representations across modalities while ignoring intra-modal training and modality-unique representations. We propose Decoupling Common and Unique Representations (DeCUR), a simple yet effective method for multimodal self-supervised learning. By distinguishing inter- and intra-modal embeddings through multimodal redundancy reduction, DeCUR can integrate complementary information across different modalities. We evaluate DeCUR in three common multimodal scenarios (radar-optical, RGB-elevation, and RGB-depth), and demonstrate its consistent improvement regardless of architectures and for both multimodal and modality-missing settings. With thorough experiments and comprehensive analysis, we hope this work can provide valuable insights and raise more interest in researching the hidden relationships of multimodal representations.

Sining Chen, Yilei Shi, Zhitong Xiong et al.

3D geo-information is of great significance for understanding the living environment; however, 3D perception from remote sensing data, especially on a large scale, is restricted. To tackle this problem, we propose a method for monocular height estimation from optical imagery, which is currently one of the richest sources of remote sensing data. As an ill-posed problem, monocular height estimation requires well-designed networks for enhanced representations to improve performance. Moreover, the distribution of height values is long-tailed with the low-height pixels, e.g., the background, as the head, and thus trained networks are usually biased and tend to underestimate building heights. To solve the problems, instead of formalizing the problem as a regression task, we propose HTC-DC Net following the classification-regression paradigm, with the head-tail cut (HTC) and the distribution-based constraints (DCs) as the main contributions. HTC-DC Net is composed of the backbone network as the feature extractor, the HTC-AdaBins module, and the hybrid regression process. The HTC-AdaBins module serves as the classification phase to determine bins adaptive to each input image. It is equipped with a vision transformer encoder to incorporate local context with holistic information and involves an HTC to address the long-tailed problem in monocular height estimation for balancing the performances of foreground and background pixels. The hybrid regression process does the regression via the smoothing of bins from the classification phase, which is trained via DCs. The proposed network is tested on three datasets of different resolutions, namely ISPRS Vaihingen (0.09 m), DFC19 (1.3 m) and GBH (3 m). Experimental results show the superiority of the proposed network over existing methods by large margins. Extensive ablation studies demonstrate the effectiveness of each design component.

Zhitong Xiong, Fahong Zhang, Yi Wang et al.

Earth observation (EO), aiming at monitoring the state of planet Earth using remote sensing data, is critical for improving our daily lives and living environment. With a growing number of satellites in orbit, an increasing number of datasets with diverse sensors and research domains are being published to facilitate the research of the remote sensing community. This paper presents a comprehensive review of more than 500 publicly published datasets, including research domains like agriculture, land use and land cover, disaster monitoring, scene understanding, vision-language models, foundation models, climate change, and weather forecasting. We systematically analyze these EO datasets from four aspects: volume, resolution distributions, research domains, and the correlation between datasets. Based on the dataset attributes, we propose to measure, rank, and select datasets to build a new benchmark for model evaluation. Furthermore, a new platform for EO, termed EarthNets, is released to achieve a fair and consistent evaluation of deep learning methods on remote sensing data. EarthNets supports standard dataset libraries and cutting-edge deep learning models to bridge the gap between the remote sensing and machine learning communities. Based on this platform, extensive deep-learning methods are evaluated on the new benchmark. The insightful results are beneficial to future research. The platform and dataset collections are publicly available at https://earthnets.github.io.

Zhitong Xiong, Haopeng Li, Xiao Xiang Zhu

Training semantic segmentation models with few annotated samples has great potential in various real-world applications. For the few-shot segmentation task, the main challenge is how to accurately measure the semantic correspondence between the support and query samples with limited training data. To address this problem, we propose to aggregate the learnable covariance matrices with a deformable 4D Transformer to effectively predict the segmentation map. Specifically, in this work, we first devise a novel hard example mining mechanism to learn covariance kernels for the Gaussian process. The learned covariance kernel functions have great advantages over existing cosine similarity-based methods in correspondence measurement. Based on the learned covariance kernels, an efficient doubly deformable 4D Transformer module is designed to adaptively aggregate feature similarity maps into segmentation results. By combining these two designs, the proposed method can not only set new state-of-the-art performance on public benchmarks, but also converge extremely faster than existing methods. Experiments on three public datasets have demonstrated the effectiveness of our method.

Yuan Yuan, Yang Zhan, Zhitong Xiong

Vision-and-language pre-training (VLP) models have experienced a surge in popularity recently. By fine-tuning them on specific datasets, significant performance improvements have been observed in various tasks. However, full fine-tuning of VLP models not only consumes a significant amount of computational resources but also has a significant environmental impact. Moreover, as remote sensing (RS) data is constantly being updated, full fine-tuning may not be practical for real-world applications. To address this issue, in this work, we investigate the parameter-efficient transfer learning (PETL) method to effectively and efficiently transfer visual-language knowledge from the natural domain to the RS domain on the image-text retrieval task. To this end, we make the following contributions. 1) We construct a novel and sophisticated PETL framework for the RS image-text retrieval (RSITR) task, which includes the pretrained CLIP model, a multimodal remote sensing adapter, and a hybrid multi-modal contrastive (HMMC) learning objective; 2) To deal with the problem of high intra-modal similarity in RS data, we design a simple yet effective HMMC loss; 3) We provide comprehensive empirical studies for PETL-based RS image-text retrieval. Our results demonstrate that the proposed method is promising and of great potential for practical applications. 4) We benchmark extensive state-of-the-art PETL methods on the RSITR task. Our proposed model only contains 0.16M training parameters, which can achieve a parameter reduction of 98.9% compared to full fine-tuning, resulting in substantial savings in training costs. Our retrieval performance exceeds traditional methods by 7-13% and achieves comparable or better performance than full fine-tuning. This work can provide new ideas and useful insights for RS vision-language tasks.

Shan Zhao, Sudipan Saha, Zhitong Xiong et al.

Precipitation nowcasting (up to a few hours) remains a challenge due to the highly complex local interactions that need to be captured accurately. Convolutional Neural Networks rely on convolutional kernels convolving with grid data and the extracted features are trapped by limited receptive field, typically expressed in excessively smooth output compared to ground truth. Thus they lack the capacity to model complex spatial relationships among the grids. Geometric deep learning aims to generalize neural network models to non-Euclidean domains. Such models are more flexible in defining nodes and edges and can effectively capture dynamic spatial relationship among geographical grids. Motivated by this, we explore a geometric deep learning-based temporal Graph Convolutional Network (GCN) for precipitation nowcasting. The adjacency matrix that simulates the interactions among grid cells is learned automatically by minimizing the L1 loss between prediction and ground truth pixel value during the training procedure. Then, the spatial relationship is refined by GCN layers while the temporal information is extracted by 1D convolution with various kernel lengths. The neighboring information is fed as auxiliary input layers to improve the final result. We test the model on sequences of radar reflectivity maps over the Trento/Italy area. The results show that GCNs improves the effectiveness of modeling the local details of the cloud profile as well as the prediction accuracy by achieving decreased error measures.

Zhitong Xiong, Yanfeng Liu, Qi Wang et al.

We present the RSSOD-Bench dataset for salient object detection (SOD) in optical remote sensing imagery. While SOD has achieved success in natural scene images with deep learning, research in SOD for remote sensing imagery (RSSOD) is still in its early stages. Existing RSSOD datasets have limitations in terms of scale, and scene categories, which make them misaligned with real-world applications. To address these shortcomings, we construct the RSSOD-Bench dataset, which contains images from four different cities in the USA. The dataset provides annotations for various salient object categories, such as buildings, lakes, rivers, highways, bridges, aircraft, ships, athletic fields, and more. The salient objects in RSSOD-Bench exhibit large-scale variations, cluttered backgrounds, and different seasons. Unlike existing datasets, RSSOD-Bench offers uniform distribution across scene categories. We benchmark 23 different state-of-the-art approaches from both the computer vision and remote sensing communities. Experimental results demonstrate that more research efforts are required for the RSSOD task.

Zhitong Xiong, Yi Wang, Fahong Zhang et al.

Earth observation (EO) in open-world settings presents a unique challenge: different applications rely on diverse sensor modalities, each with varying ground sampling distances, spectral ranges, and numbers of spectral bands. However, existing EO foundation models are typically tailored to specific sensor types, making them inflexible when generalizing across the heterogeneous landscape of EO data. To address this, we propose the Dynamic One-For-All (DOFA) model, a unified, multimodal foundation framework designed for diverse vision tasks in EO. Inspired by neural plasticity, DOFA utilizes a wavelength-conditioned dynamic hypernetwork to process inputs from five distinct satellite sensors flexibly. By continually pretraining on five EO modalities, DOFA achieves state-of-the-art performance across multiple downstream tasks and generalizes well to unseen modalities. Enhanced with hybrid continual pretraining, DOFA+ requires significantly fewer computational resources while outperforming counterparts trained with extensive GPU budgets. Experiments on diverse datasets highlight DOFA's potential as a foundation for general-purpose vision models in the sensor-diverse EO domain. The code and pre-trained weights are publicly available at https://github.com/zhu-xlab/DOFA.

Yang Zhan, Yuan Yuan, Zhitong Xiong

We introduce a novel task of 3D visual grounding in monocular RGB images using language descriptions with both appearance and geometry information. Specifically, we build a large-scale dataset, Mono3DRefer, which contains 3D object targets with their corresponding geometric text descriptions, generated by ChatGPT and refined manually. To foster this task, we propose Mono3DVG-TR, an end-to-end transformer-based network, which takes advantage of both the appearance and geometry information in text embeddings for multi-modal learning and 3D object localization. Depth predictor is designed to explicitly learn geometry features. The dual text-guided adapter is proposed to refine multiscale visual and geometry features of the referred object. Based on depth-text-visual stacking attention, the decoder fuses object-level geometric cues and visual appearance into a learnable query. Comprehensive benchmarks and some insightful analyses are provided for Mono3DVG. Extensive comparisons and ablation studies show that our method significantly outperforms all baselines. The dataset and code will be publicly available at: https://github.com/ZhanYang-nwpu/Mono3DVG.

Nils Lehmann, Yi Wang, Zhitong Xiong et al.

State-of-the-art generative image and video models rely heavily on tokenizers that compress high-dimensional inputs into more efficient latent representations. While this paradigm has revolutionized RGB generation, Earth observation (EO) data presents unique challenges due to diverse sensor specifications and variable spectral channels. We propose EO-VAE, a multi-sensor variational autoencoder designed to serve as a foundational tokenizer for the EO domain. Unlike prior approaches that train separate tokenizers for each modality, EO-VAE utilizes a single model to encode and reconstruct flexible channel combinations via dynamic hypernetworks. Our experiments on the TerraMesh dataset demonstrate that EO-VAE achieves superior reconstruction fidelity compared to the TerraMind tokenizers, establishing a robust baseline for latent generative modeling in remote sensing.

Yi Wang, Zhitong Xiong, Chenying Liu et al.

Advances in Earth observation (EO) foundation models have unlocked the potential of big satellite data to learn generic representations from space, benefiting a wide range of downstream applications crucial to our planet. However, most existing efforts remain limited to fixed spectral sensors, focus solely on the Earth's surface, and overlook valuable metadata beyond imagery. In this work, we take a step towards next-generation EO foundation models with three key components: 1) Copernicus-Pretrain, a massive-scale pretraining dataset that integrates 18.7M aligned images from all major Copernicus Sentinel missions, spanning from the Earth's surface to its atmosphere; 2) Copernicus-FM, a unified foundation model capable of processing any spectral or non-spectral sensor modality using extended dynamic hypernetworks and flexible metadata encoding; and 3) Copernicus-Bench, a systematic evaluation benchmark with 15 hierarchical downstream tasks ranging from preprocessing to specialized applications for each Sentinel mission. Our dataset, model, and benchmark greatly improve the scalability, versatility, and multimodal adaptability of EO foundation models, while also creating new opportunities to connect EO, weather, and climate research. Codes, datasets and models are available at https://github.com/zhu-xlab/Copernicus-FM.

Xuran Hu, Zhitong Xiong, Zhongcheng Hong et al.

Current Large Multimodal Models (LMMs) in Earth Observation typically neglect the critical "vertical" dimension, limiting their reasoning capabilities in complex remote sensing geometries and disaster scenarios where physical spatial structures often outweigh planar visual textures. To bridge this gap, we introduce a comprehensive evaluation framework dedicated to height-aware remote sensing understanding. First, to overcome the severe scarcity of annotated data, we develop a scalable, VLM-driven data generation pipeline utilizing systematic prompt engineering and metadata extraction. This pipeline constructs two complementary benchmarks: GeoHeight-Bench for relative height analysis, and a more challenging GeoHeight-Bench+ for holistic, terrain-aware reasoning. Furthermore, to validate the necessity of height perception, we propose GeoHeightChat, the first height-aware remote sensing LMM baseline. Serving as a strong proof of concept, our baseline demonstrates that synergizing visual semantics with implicitly injected height geometric features effectively mitigates the "vertical blind spot", successfully unlocking a new paradigm of interactive height reasoning in existing optical models.

Xiang Li, Yong Tao, Siyuan Zhang et al.

Earth observation foundation models have shown strong generalization across multiple Earth observation tasks, but their robustness under real-world perturbations remains underexplored. To bridge this gap, we introduce REOBench, the first comprehensive benchmark for evaluating the robustness of Earth observation foundation models across six tasks and twelve types of image corruptions, including both appearance-based and geometric perturbations. To ensure realistic and fine-grained evaluation, our benchmark focuses on high-resolution optical remote sensing images, which are widely used in critical applications such as urban planning and disaster response. We conduct a systematic evaluation of a broad range of models trained using masked image modeling, contrastive learning, and vision-language pre-training paradigms. Our results reveal that (1) existing Earth observation foundation models experience significant performance degradation when exposed to input corruptions. (2) The severity of degradation varies across tasks, model architectures, backbone sizes, and types of corruption, with performance drop varying from less than 1% to over 20%. (3) Vision-language models show enhanced robustness, particularly in multimodal tasks. REOBench underscores the vulnerability of current Earth observation foundation models to real-world corruptions and provides actionable insights for developing more robust and reliable models. Code and data are publicly available at https://github.com/lx709/REOBench.

Yan Shu, Bin Ren, Zhitong Xiong et al.

Vision-language models (VLMs) have shown promise in earth observation (EO), yet they struggle with tasks that require grounding complex spatial reasoning in precise pixel-level visual representations. To address this problem, we introduce TerraScope, a unified VLM that delivers pixel-grounded geospatial reasoning with two key capabilities: (1) modality-flexible reasoning: it handles single-modality inputs (optical or SAR) and adaptively fuses different modalities into the reasoning process when both are available; (2) multi-temporal reasoning: it integrates temporal sequences for change analysis across multiple time points. In addition, we curate Terra-CoT, a large-scale dataset containing 1 million samples with pixel-level masks embedded in reasoning chains across multiple sources. We also propose TerraScope-Bench, the first benchmark for pixel-grounded geospatial reasoning with six sub-tasks that evaluates both answer accuracy and mask quality to ensure authentic pixel-grounded reasoning. Experiments show that TerraScope significantly outperforms existing VLMs on pixel-grounded geospatial reasoning while providing interpretable visual evidence.

Shan Zhao, Zhitong Xiong, Jie Zhao et al.

Our planet is facing increasingly frequent extreme events, which pose major risks to human lives and ecosystems. Recent advances in machine learning (ML), especially with foundation models (FMs) trained on extensive datasets, excel in extracting features and show promise in disaster management. Nevertheless, these models often inherit biases from training data, challenging their performance over extreme values. To explore the reliability of FM in the context of extreme events, we introduce \textbf{ExE}Bench (\textbf{Ex}treme \textbf{E}arth Benchmark), a collection of seven extreme event categories across floods, wildfires, storms, tropical cyclones, extreme precipitation, heatwaves, and cold waves. The dataset features global coverage, varying data volumes, and diverse data sources with different spatial, temporal, and spectral characteristics. To broaden the real-world impact of FMs, we include multiple challenging ML tasks that are closely aligned with operational needs in extreme events detection, monitoring, and forecasting. ExEBench aims to (1) assess FM generalizability across diverse, high-impact tasks and domains, (2) promote the development of novel ML methods that benefit disaster management, and (3) offer a platform for analyzing the interactions and cascading effects of extreme events to advance our understanding of Earth system, especially under the climate change expected in the decades to come. The dataset and code are public https://github.com/zhaoshan2/EarthExtreme-Bench.

Yang Zhan, Zhitong Xiong, Yuan Yuan

Large language models (LLMs) have recently been extended to the vision-language realm, obtaining impressive general multi-modal capabilities. However, the exploration of multi-modal large language models (MLLMs) for remote sensing (RS) data is still in its infancy, and the performance is not satisfactory. In this work, we introduce SkyEyeGPT, a unified multi-modal large language model specifically designed for RS vision-language understanding. To this end, we meticulously curate an RS multi-modal instruction tuning dataset, including single-task and multi-task conversation instructions. After manual verification, we obtain a high-quality RS instruction-following dataset with 968k samples. Our research demonstrates that with a simple yet effective design, SkyEyeGPT works surprisingly well on considerably different tasks without the need for extra encoding modules. Specifically, after projecting RS visual features to the language domain via an alignment layer, they are fed jointly with task-specific instructions into an LLM-based RS decoder to predict answers for RS open-ended tasks. In addition, we design a two-stage tuning method to enhance instruction-following and multi-turn dialogue ability at different granularities. Experiments on 8 datasets for RS vision-language tasks demonstrate SkyEyeGPT's superiority in image-level and region-level tasks, such as captioning and visual grounding. In particular, SkyEyeGPT exhibits encouraging results compared to GPT-4V in some qualitative tests. The online demo, code, and dataset will be released in https://github.com/ZhanYang-nwpu/SkyEyeGPT.

Zhitong Xiong, Sining Chen, Yi Wang et al.

Geometric information in the normalized digital surface models (nDSM) is highly correlated with the semantic class of the land cover. Exploiting two modalities (RGB and nDSM (height)) jointly has great potential to improve the segmentation performance. However, it is still an under-explored field in remote sensing due to the following challenges. First, the scales of existing datasets are relatively small and the diversity of existing datasets is limited, which restricts the ability of validation. Second, there is a lack of unified benchmarks for performance assessment, which leads to difficulties in comparing the effectiveness of different models. Last, sophisticated multi-modal semantic segmentation methods have not been deeply explored for remote sensing data. To cope with these challenges, in this paper, we introduce a new remote-sensing benchmark dataset for multi-modal semantic segmentation based on RGB-Height (RGB-H) data. Towards a fair and comprehensive analysis of existing methods, the proposed benchmark consists of 1) a large-scale dataset including co-registered RGB and nDSM pairs and pixel-wise semantic labels; 2) a comprehensive evaluation and analysis of existing multi-modal fusion strategies for both convolutional and Transformer-based networks on remote sensing data. Furthermore, we propose a novel and effective Transformer-based intermediary multi-modal fusion (TIMF) module to improve the semantic segmentation performance through adaptive token-level multi-modal fusion.The designed benchmark can foster future research on developing new methods for multi-modal learning on remote sensing data. Extensive analyses of those methods are conducted and valuable insights are provided through the experimental results. Code for the benchmark and baselines can be accessed at \url{https://github.com/EarthNets/RSI-MMSegmentation}.

Xiao Xiang Zhu, Zhitong Xiong, Yi Wang et al.

Foundation models have enormous potential in advancing Earth and climate sciences, however, current approaches may not be optimal as they focus on a few basic features of a desirable Earth and climate foundation model. Crafting the ideal Earth foundation model, we define eleven features which would allow such a foundation model to be beneficial for any geoscientific downstream application in an environmental- and human-centric manner.We further shed light on the way forward to achieve the ideal model and to evaluate Earth foundation models. What comes after foundation models? Energy efficient adaptation, adversarial defenses, and interpretability are among the emerging directions.

Zhitong Xiong, Yi Wang, Fahong Zhang et al.

Foundation models characterized by extensive parameters and trained on large-scale datasets have demonstrated remarkable efficacy across various downstream tasks for remote sensing data. Current remote sensing foundation models typically specialize in a single modality or a specific spatial resolution range, limiting their versatility for downstream datasets. While there have been attempts to develop multi-modal remote sensing foundation models, they typically employ separate vision encoders for each modality or spatial resolution, necessitating a switch in backbones contingent upon the input data. To address this issue, we introduce a simple yet effective method, termed OFA-Net (One-For-All Network): employing a single, shared Transformer backbone for multiple data modalities with different spatial resolutions. Using the masked image modeling mechanism, we pre-train a single Transformer backbone on a curated multi-modal dataset with this simple design. Then the backbone model can be used in different downstream tasks, thus forging a path towards a unified foundation backbone model in Earth vision. The proposed method is evaluated on 12 distinct downstream tasks and demonstrates promising performance.

Zhenghang Yuan, Zhitong Xiong, Lichao Mou et al.

An in-depth comprehension of global land cover is essential in Earth observation, forming the foundation for a multitude of applications. Although remote sensing technology has advanced rapidly, leading to a proliferation of satellite imagery, the inherent complexity of these images often makes them difficult for non-expert users to understand. Natural language, as a carrier of human knowledge, can be a bridge between common users and complicated satellite imagery. In this context, we introduce a global-scale, high-quality image-text dataset for remote sensing, providing natural language descriptions for Sentinel-2 data to facilitate the understanding of satellite imagery for common users. Specifically, we utilize Sentinel-2 data for its global coverage as the foundational image source, employing semantic segmentation labels from the European Space Agency's (ESA) WorldCover project to enrich the descriptions of land covers. By conducting in-depth semantic analysis, we formulate detailed prompts to elicit rich descriptions from ChatGPT. To enhance the dataset's quality, we introduce the manual verification process. This step involves manual inspection and correction to refine the dataset, thus significantly improving its accuracy and quality. Finally, we offer the community ChatEarthNet, a large-scale image-text dataset characterized by global coverage, high quality, wide-ranging diversity, and detailed descriptions. ChatEarthNet consists of 163,488 image-text pairs with captions generated by ChatGPT-3.5 and an additional 10,000 image-text pairs with captions generated by ChatGPT-4V(ision). This dataset has significant potential for training vision-language geo-foundation models and evaluating large vision-language models for remote sensing. The dataset will be made publicly available.

Shan Zhao, Zhaiyu Chen, Zhitong Xiong et al.

Earth Observation (EO) data analysis has been significantly revolutionized by deep learning (DL), with applications typically limited to grid-like data structures. Graph Neural Networks (GNNs) emerge as an important innovation, propelling DL into the non-Euclidean domain. Naturally, GNNs can effectively tackle the challenges posed by diverse modalities, multiple sensors, and the heterogeneous nature of EO data. To introduce GNNs in the related domains, our review begins by offering fundamental knowledge on GNNs. Then, we summarize the generic problems in EO, to which GNNs can offer potential solutions. Following this, we explore a broad spectrum of GNNs' applications to scientific problems in Earth systems, covering areas such as weather and climate analysis, disaster management, air quality monitoring, agriculture, land cover classification, hydrological process modeling, and urban modeling. The rationale behind adopting GNNs in these fields is explained, alongside methodologies for organizing graphs and designing favorable architectures for various tasks. Furthermore, we highlight methodological challenges of implementing GNNs in these domains and possible solutions that could guide future research. While acknowledging that GNNs are not a universal solution, we conclude the paper by comparing them with other popular architectures like transformers and analyzing their potential synergies.

Leonard Waldmann, Ando Shah, Yi Wang et al.

Earth observation (EO) data features diverse sensing platforms with varying spectral bands, spatial resolutions, and sensing modalities. While most prior work has constrained inputs to fixed sensors, a new class of any-sensor foundation models able to process arbitrary sensors has recently emerged. Contributing to this line of work, we propose Panopticon, an any-sensor foundation model built on the DINOv2 framework. We extend DINOv2 by (1) treating images of the same geolocation across sensors as natural augmentations, (2) subsampling channels to diversify spectral input, and (3) adding a cross attention over channels as a flexible patch embedding mechanism. By encoding the wavelength and modes of optical and synthetic aperture radar sensors, respectively, Panopticon can effectively process any combination of arbitrary channels. In extensive evaluations, we achieve state-of-the-art performance on GEO-Bench, especially on the widely-used Sentinel-1 and Sentinel-2 sensors, while out-competing other any-sensor models, as well as domain adapted fixed-sensor models on unique sensor configurations. Panopticon enables immediate generalization to both existing and future satellite platforms, advancing sensor-agnostic EO.

Shan Zhao, Ioannis Prapas, Ilektra Karasante et al.

Wildfire forecasting is notoriously hard due to the complex interplay of different factors such as weather conditions, vegetation types and human activities. Deep learning models show promise in dealing with this complexity by learning directly from data. However, to inform critical decision making, we argue that we need models that are right for the right reasons; that is, the implicit rules learned should be grounded by the underlying processes driving wildfires. In that direction, we propose integrating causality with Graph Neural Networks (GNNs) that explicitly model the causal mechanism among complex variables via graph learning. The causal adjacency matrix considers the synergistic effect among variables and removes the spurious links from highly correlated impacts. Our methodology's effectiveness is demonstrated through superior performance forecasting wildfire patterns in the European boreal and mediterranean biome. The gain is especially prominent in a highly imbalanced dataset, showcasing an enhanced robustness of the model to adapt to regime shifts in functional relationships. Furthermore, SHAP values from our trained model further enhance our understanding of the model's inner workings.

Zhitong Xiong, Yi Wang, Weikang Yu et al.

Earth observation (EO) spans a broad spectrum of modalities, including optical, radar, multispectral, and hyperspectral data, each capturing distinct environmental signals. However, current vision-language models in EO, particularly CLIP-based variants, remain confined to individual modalities, limiting generalization and scalability across diverse tasks. We present DOFA-CLIP (Dynamic-One-For-All CLIP), a unified vision-language foundation model that dynamically adapts to EO modalities with flexible spectral configurations through a single Transformer backbone. Our approach introduces three key contributions: 1) the construction of GeoLangBind-2M, a large-scale EO image-text dataset covering six heterogeneous modalities with rich natural language descriptions; 2) a novel training strategy called VECT (Vision-models Enhanced Contrastive Text-image pretraining), which enhances the spatial awareness of CLIP features with multiple vision foundation models; and 3) a Modality-aware Knowledge Agglomeration (MaKA) module that refines feature distillation with modality-specific awareness. DOFA-CLIP achieves state-of-the-art zero-shot performance across a wide range of EO benchmarks, including unseen modalities and a diverse number of input spectral bands. Together, these contributions establish a scalable foundation for multimodal EO understanding and open new avenues for integrating heterogeneous EO data with large language models. Code and datasets will be released. Code and datasets are publicly available.

Shan Zhao, Zhitong Xiong, Xiao Xiang Zhu

Subseasonal forecasting, which is pivotal for agriculture, water resource management, and early warning of disasters, faces challenges due to the chaotic nature of the atmosphere. Recent advances in machine learning (ML) have revolutionized weather forecasting by achieving competitive predictive skills to numerical models. However, training such foundation models requires thousands of GPU days, which causes substantial carbon emissions and limits their broader applicability. Moreover, ML models tend to fool the pixel-wise error scores by producing smoothed results which lack physical consistency and meteorological meaning. To deal with the aforementioned problems, we propose a teleconnection-informed transformer. Our architecture leverages the pretrained Pangu model to achieve good initial weights and integrates a teleconnection-informed temporal module to improve predictability in an extended temporal range. Remarkably, by adjusting 1.1% of the Pangu model's parameters, our method enhances predictability on four surface and five upper-level atmospheric variables at a two-week lead time. Furthermore, the teleconnection-filtered features improve the spatial granularity of outputs significantly, indicating their potential physical consistency. Our research underscores the importance of atmospheric and oceanic teleconnections in driving future weather conditions. Besides, it presents a resource-efficient pathway for researchers to leverage existing foundation models on versatile downstream tasks.

Wei Huang, Zhitong Xiong, Chenying Liu et al.

The performance of deep learning models in remote sensing (RS) strongly depends on the availability of high-quality labeled data. However, collecting large-scale annotations is costly and time-consuming, while vast amounts of unlabeled imagery remain underutilized. To address this challenge, we propose a Hierarchical Semi-Supervised Active Learning (HSSAL) framework that integrates semi-supervised learning (SSL) and a novel hierarchical active learning (HAL) in a closed iterative loop. In each iteration, SSL refines the model using both labeled data through supervised learning and unlabeled data via weak-to-strong self-training, improving feature representation and uncertainty estimation. Guided by the refined representations and uncertainty cues of unlabeled samples, HAL then conducts sample querying through a progressive clustering strategy, selecting the most informative instances that jointly satisfy the criteria of scalability, diversity, and uncertainty. This hierarchical process ensures both efficiency and representativeness in sample selection. Extensive experiments on three benchmark RS scene classification datasets, including UCM, AID, and NWPU-RESISC45, demonstrate that HSSAL consistently outperforms SSL- or AL-only baselines. Remarkably, with only 8%, 4%, and 2% labeled training data on UCM, AID, and NWPU-RESISC45, respectively, HSSAL achieves over 95% of fully-supervised accuracy, highlighting its superior label efficiency through informativeness exploitation of unlabeled data. Our code will be publicly available.

Yan Shu, Bin Ren, Zhitong Xiong et al.

Earth Observation (EO) data analysis is vital for monitoring environmental and human dynamics. Recent Multimodal Large Language Models (MLLMs) show potential in EO understanding but remain restricted to single-sensor inputs, overlooking the complementarity across heterogeneous modalities. We propose EarthMind, a unified vision-language framework that handles both single- and cross-sensor inputs via an innovative hierarchical cross-modal attention (ie, HCA) design. Specifically, HCA hierarchically captures visual relationships across sensors and aligns them with language queries, enabling adaptive fusion of optical and Synthetic Aperture Radar (SAR) features. To support cross-sensor learning, we curate FusionEO, a 30K-pair dataset with diverse annotations, and establish EarthMind-Bench, a 2,841-pair benchmark with expert annotations for perception and reasoning tasks. Extensive experiments show that EarthMind achieves state-of-the-art results on EarthMind-Bench and surpasses existing MLLMs on multiple EO benchmarks.

Yang Mu, Zhitong Xiong, Yi Wang et al.

Global tree species mapping using remote sensing data is vital for biodiversity monitoring, forest management, and ecological research. However, progress in this field has been constrained by the scarcity of large-scale, labeled datasets. To address this, we introduce GlobalGeoTree, a comprehensive global dataset for tree species classification. GlobalGeoTree comprises 6.3 million geolocated tree occurrences, spanning 275 families, 2,734 genera, and 21,001 species across the hierarchical taxonomic levels. Each sample is paired with Sentinel-2 image time series and 27 auxiliary environmental variables, encompassing bioclimatic, geographic, and soil data. The dataset is partitioned into GlobalGeoTree-6M for model pretraining and curated evaluation subsets, primarily GlobalGeoTree-10kEval for zero-shot and few-shot benchmarking. To demonstrate the utility of the dataset, we introduce a baseline model, GeoTreeCLIP, which leverages paired remote sensing data and taxonomic text labels within a vision-language framework pretrained on GlobalGeoTree-6M. Experimental results show that GeoTreeCLIP achieves substantial improvements in zero- and few-shot classification on GlobalGeoTree-10kEval over existing advanced models. By making the dataset, models, and code publicly available, we aim to establish a benchmark to advance tree species classification and foster innovation in biodiversity research and ecological applications.

Jie Zhao, Zhitong Xiong, Xiao Xiang Zhu

Due to its cloud-penetrating capability and independence from solar illumination, satellite Synthetic Aperture Radar (SAR) is the preferred data source for large-scale flood mapping, providing global coverage and including various land cover classes. However, most studies on large-scale SAR-derived flood mapping using deep learning algorithms have primarily focused on flooded open areas, utilizing available open-access datasets (e.g., Sen1Floods11) and with limited attention to urban floods. To address this gap, we introduce \textbf{UrbanSARFloods}, a floodwater dataset featuring pre-processed Sentinel-1 intensity data and interferometric coherence imagery acquired before and during flood events. It contains 8,879 $512\times 512$ chips covering 807,500 $km^2$ across 20 land cover classes and 5 continents, spanning 18 flood events. We used UrbanSARFloods to benchmark existing state-of-the-art convolutional neural networks (CNNs) for segmenting open and urban flood areas. Our findings indicate that prevalent approaches, including the Weighted Cross-Entropy (WCE) loss and the application of transfer learning with pretrained models, fall short in overcoming the obstacles posed by imbalanced data and the constraints of a small training dataset. Urban flood detection remains challenging. Future research should explore strategies for addressing imbalanced data challenges and investigate transfer learning's potential for SAR-based large-scale flood mapping. Besides, expanding this dataset to include additional flood events holds promise for enhancing its utility and contributing to advancements in flood mapping techniques.

Zhitong Xiong, Sining Chen, Yilei Shi et al.

Monocular height estimation (MHE) from remote sensing imagery has high potential in generating 3D city models efficiently for a quick response to natural disasters. Most existing works pursue higher performance. However, there is little research exploring the interpretability of MHE networks. In this paper, we target at exploring how deep neural networks predict height from a single monocular image. Towards a comprehensive understanding of MHE networks, we propose to interpret them from multiple levels: 1) Neurons: unit-level dissection. Exploring the semantic and height selectivity of the learned internal deep representations; 2) Instances: object-level interpretation. Studying the effects of different semantic classes, scales, and spatial contexts on height estimation; 3) Attribution: pixel-level analysis. Understanding which input pixels are important for the height estimation. Based on the multi-level interpretation, a disentangled latent Transformer network is proposed towards a more compact, reliable, and explainable deep model for monocular height estimation. Furthermore, a novel unsupervised semantic segmentation task based on height estimation is first introduced in this work. Additionally, we also construct a new dataset for joint semantic segmentation and height estimation. Our work provides novel insights for both understanding and designing MHE models.

Zhitong Xiong, Wei Huang, Jingtao Hu et al.

Generating 3D city models rapidly is crucial for many applications. Monocular height estimation is one of the most efficient and timely ways to obtain large-scale geometric information. However, existing works focus primarily on training and testing models using unbiased datasets, which does not align well with real-world applications. Therefore, we propose a new benchmark dataset to study the transferability of height estimation models in a cross-dataset setting. To this end, we first design and construct a large-scale benchmark dataset for cross-dataset transfer learning on the height estimation task. This benchmark dataset includes a newly proposed large-scale synthetic dataset, a newly collected real-world dataset, and four existing datasets from different cities. Next, a new experimental protocol, few-shot cross-dataset transfer, is designed. Furthermore, in this paper, we propose a scale-deformable convolution module to enhance the window-based Transformer for handling the scale-variation problem in the height estimation task. Experimental results have demonstrated the effectiveness of the proposed methods in the traditional and cross-dataset transfer settings. The datasets and codes are publicly available at https://mediatum.ub.tum.de/1662763 and https://thebenchmarkh.github.io/.

Zhenghang Yuan, Lichao Mou, Zhitong Xiong et al.

The Earth's surface is continually changing, and identifying changes plays an important role in urban planning and sustainability. Although change detection techniques have been successfully developed for many years, these techniques are still limited to experts and facilitators in related fields. In order to provide every user with flexible access to change information and help them better understand land-cover changes, we introduce a novel task: change detection-based visual question answering (CDVQA) on multi-temporal aerial images. In particular, multi-temporal images can be queried to obtain high level change-based information according to content changes between two input images. We first build a CDVQA dataset including multi-temporal image-question-answer triplets using an automatic question-answer generation method. Then, a baseline CDVQA framework is devised in this work, and it contains four parts: multi-temporal feature encoding, multi-temporal fusion, multi-modal fusion, and answer prediction. In addition, we also introduce a change enhancing module to multi-temporal feature encoding, aiming at incorporating more change-related information. Finally, effects of different backbones and multi-temporal fusion strategies are studied on the performance of CDVQA task. The experimental results provide useful insights for developing better CDVQA models, which are important for future research on this task.

Zhitong Xiong, Yuan Yuan, Qi Wang

Indoor scene images usually contain scattered objects and various scene layouts, which make RGB-D scene classification a challenging task. Existing methods still have limitations for classifying scene images with great spatial variability. Thus, how to extract local patch-level features effectively using only image labels is still an open problem for RGB-D scene recognition. In this paper, we propose an efficient framework for RGB-D scene recognition, which adaptively selects important local features to capture the great spatial variability of scene images. Specifically, we design a differentiable local feature selection (DLFS) module, which can extract the appropriate number of key local scenerelated features. Discriminative local theme-level and object-level representations can be selected with the DLFS module from the spatially-correlated multi-modal RGB-D features. We take advantage of the correlation between RGB and depth modalities to provide more cues for selecting local features. To ensure that discriminative local features are selected, the variational mutual information maximization loss is proposed. Additionally, the DLFS module can be easily extended to select local features of different scales. By concatenating the local-orderless and global structured multi-modal features, the proposed framework can achieve state-of-the-art performance on public RGB-D scene recognition datasets.

Chuang Yang, Mulin Chen, Zhitong Xiong et al.

Recently fast arbitrary-shaped text detection has become an attractive research topic. However, most existing methods are non-real-time, which may fall short in intelligent systems. Although a few real-time text methods are proposed, the detection accuracy is far behind non-real-time methods. To improve the detection accuracy and speed simultaneously, we propose a novel fast and accurate text detection framework, namely CM-Net, which is constructed based on a new text representation method and a multi-perspective feature (MPF) module. The former can fit arbitrary-shaped text contours by concentric mask (CM) in an efficient and robust way. The latter encourages the network to learn more CM-related discriminative features from multiple perspectives and brings no extra computational cost. Benefiting the advantages of CM and MPF, the proposed CM-Net only needs to predict one CM of the text instance to rebuild the text contour and achieves the best balance between detection accuracy and speed compared with previous works. Moreover, to ensure that multi-perspective features are effectively learned, the multi-factor constraints loss is proposed. Extensive experiments demonstrate the proposed CM is efficient and robust to fit arbitrary-shaped text instances, and also validate the effectiveness of MPF and constraints loss for discriminative text features recognition. Furthermore, experimental results show that the proposed CM-Net is superior to existing state-of-the-art (SOTA) real-time text detection methods in both detection speed and accuracy on MSRA-TD500, CTW1500, Total-Text, and ICDAR2015 datasets.

Yuan Yuan, Zhitong Xiong, Qi Wang

Although traffic sign detection has been studied for years and great progress has been made with the rise of deep learning technique, there are still many problems remaining to be addressed. For complicated real-world traffic scenes, there are two main challenges. Firstly, traffic signs are usually small size objects, which makes it more difficult to detect than large ones; Secondly, it is hard to distinguish false targets which resemble real traffic signs in complex street scenes without context information. To handle these problems, we propose a novel end-to-end deep learning method for traffic sign detection in complex environments. Our contributions are as follows: 1) We propose a multi-resolution feature fusion network architecture which exploits densely connected deconvolution layers with skip connections, and can learn more effective features for the small size object; 2) We frame the traffic sign detection as a spatial sequence classification and regression task, and propose a vertical spatial sequence attention (VSSA) module to gain more context information for better detection performance. To comprehensively evaluate the proposed method, we do experiments on several traffic sign datasets as well as the general object detection dataset and the results have shown the effectiveness of our proposed method.