arxiv: 2604.11415 · v1 · submitted 2026-04-13 · 💻 cs.CV

Recognition: unknown

Observe Less, Understand More: Cost-aware Cross-scale Observation for Remote Sensing Understanding

Zhenghao Xie , Jing Xiao , Zhenqi Wang , Kexin Ma , Liang Liao , Gui-Song Xia , Mi Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:13 UTC · model grok-4.3

classification 💻 cs.CV

keywords remote sensingcross-scale observationcost-aware samplinghigh-resolution imagerymulti-resolution dataimage recognitioncontent-based retrievalbenchmark dataset

0 comments

The pith

Formulating remote sensing as a cost-aware problem that couples selective high-resolution sampling with cross-patch low-resolution prediction improves task performance at lower acquisition cost.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Remote sensing applications need both broad low-resolution coverage for context and selective high-resolution details for accuracy, but high-resolution imagery is expensive to acquire and limited in extent. The paper argues that existing selection methods, which decide on isolated low-resolution patches, produce fragmented representations and miss optimal regions. By instead framing the entire process as one joint cost-aware optimization that predicts cross-patch representations while choosing where to sample high-resolution data, the approach reasons more effectively from fewer expensive observations. A new benchmark of ten million aligned multi-resolution images supports controlled testing of such budget-limited strategies. Experiments on recognition and retrieval confirm better performance per unit of high-resolution cost.

Core claim

The paper establishes that cross-scale remote sensing understanding can be cast as a single cost-aware problem in which fine-grained high-resolution sampling decisions are made jointly with cross-patch representation prediction from low-resolution imagery, yielding more effective scene reasoning under constrained high-resolution budgets than methods relying on isolated patch selections.

What carries the argument

The unified cost-aware cross-scale observation framework that couples fine-grained high-resolution sampling with cross-patch representation prediction from low-resolution data.

If this is right

Superior performance-cost trade-off on recognition and retrieval tasks compared with prior selection strategies.
The GL-10M dataset of ten million spatially aligned multi-resolution images enables systematic testing of budget-constrained cross-scale methods.
Avoidance of fragmented features and suboptimal reasoning that arise when high-resolution patches are chosen without cross-patch context.
Efficient global low-resolution observation combined with targeted high-resolution acquisition for improved overall task results.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same coupling of cheap global prediction with selective expensive sampling could be tested in other multi-resolution domains such as medical imaging or autonomous driving to reduce sensor costs.
If cross-patch prediction proves stable, the method suggests a path toward real-time adaptive acquisition policies on orbiting platforms where downlink or storage budgets are limited.
Extending the framework to sequences of images rather than single scenes would allow testing whether temporal context further reduces the number of high-resolution frames needed.
Comparing the approach against reinforcement-learning-based active sensing policies on the same benchmark would clarify whether the joint formulation offers advantages over sequential decision models.

Load-bearing premise

That predictions of cross-patch representations from low-resolution imagery can reliably indicate which high-resolution patches contain the critical task-relevant details without overlooking important local information.

What would settle it

On the GL-10M benchmark or similar data, run the method at a fixed high-resolution budget and observe whether task accuracy falls below that of uniform or random high-resolution sampling at the same total cost, or whether selected patches systematically miss key local features.

Figures

Figures reproduced from arXiv: 2604.11415 by Gui-Song Xia, Jing Xiao, Kexin Ma, Liang Liao, Mi Wang, Zhenghao Xie, Zhenqi Wang.

**Figure 1.** Figure 1: Overview of remote sensing understanding under limited HR observation. (a) Dense HR observations incur high [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Figure 2: Cost-aware cross-scale understanding with two modules and latent alignment. Given the LR global overview [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Qualitative visualization of the sampler supervision [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Robustness of sparse observation under different [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: Qualitative comparison of latent completion by the observation-guided latent predictor under sparse HR observation. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Label-frequency distribution of GL-10M. We report the frequency of all 66 labels in the combined train and validation [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗

**Figure 7.** Figure 7: Hallucination analysis with and without LR guidance. From left to right: GT, w/o LR guidance, with LR guidance, and [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗

**Figure 8.** Figure 8: Qualitative budget sweep under threshold-controlled HR acquisition. We sweep the threshold from 0.00 (full HR, GT) [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗

**Figure 9.** Figure 9: Representative failure cases. (a) missed informative [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗

read the original abstract

Remote sensing understanding inherently requires multi-resolution observation, since different targets and application tasks demand different levels of spatial detail. While low-resolution (LR) imagery enables efficient global observation, high-resolution (HR) imagery provides critical local details at much higher acquisition cost and limited coverage. This motivates a cross-scale sensing strategy that selectively acquires HR imagery from LR-based global perception to improve task performance under constrained cost. Existing methods for HR sampling methods typically make selection decisions from isolated LR patches, which ignore fine-grained intra-patch importance and cross-patch contextual interactions, leading to fragmented feature representation and suboptimal scene reasoning under sparse HR observations. To address this issue, we formulate cross-scale remote sensing understanding as a unified cost-aware problem that couples fine-grained HR sampling with cross-patch representation prediction, enabling more effective task reasoning with fewer HR observations. Furthermore, we present GL-10M, a large-scale benchmark of 10 million spatially aligned multi-resolution images, enabling systematic evaluation of budget-constrained cross-scale reasoning in remote sensing. Extensive experiments on recognition and retrieval tasks show that our method consistently achieves a superior performance-cost trade-off.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper unifies HR patch sampling with cross-patch representation prediction under a cost budget and ships a 10M-image multi-res benchmark, but the performance edge is stated without numbers or ablations in the abstract.

read the letter

The main takeaway is that they treat sampling and scene understanding as one coupled optimization instead of two separate stages. They also introduce GL-10M, a large set of spatially aligned low- and high-resolution images meant for testing budget-constrained remote-sensing tasks. That combination is the concrete addition here. The formulation tries to fix the problem of isolated low-res decisions by letting the model predict how patches relate to each other and what representations they should carry. This could reduce the chance of grabbing irrelevant high-res patches when global context matters. The benchmark itself is a practical step forward; anyone running recognition or retrieval experiments under cost limits can use the same data to compare approaches. The experiments are said to deliver a better performance-cost curve on those tasks, yet the abstract supplies no accuracy deltas, no cost ratios, no error bars, and no ablation on the cross-patch term. Without those, it is hard to judge whether the coupling actually delivers the claimed advantage or simply moves the selection problem elsewhere. The weakest link remains the assumption that low-resolution context plus the prediction head can surface the fine local details that matter most. If small objects or textures are decisive but invisible at low resolution, the model may still allocate the high-res budget to the wrong places. This is aimed at remote-sensing groups that already manage multi-scale acquisition and want to tighten their budgets. A reader working on active sampling or multi-resolution fusion could pick up the benchmark and the joint formulation even if they keep their own predictor. I would send it to peer review. The dataset is large enough to be worth checking, and the formulation is clear enough that referees can test the central assumption directly against the numbers.

Referee Report

2 major / 1 minor

Summary. The paper formulates cross-scale remote sensing understanding as a unified cost-aware problem that couples fine-grained HR sampling decisions with cross-patch representation prediction, in contrast to prior methods that select from isolated LR patches. It introduces the GL-10M benchmark of 10 million spatially aligned multi-resolution images and claims that extensive experiments on recognition and retrieval tasks demonstrate a consistently superior performance-cost trade-off.

Significance. If the central claim holds, the work could improve efficiency in remote sensing applications where HR acquisition is costly and coverage-limited, by enabling more informed selection of sparse HR patches guided by cross-patch context. The GL-10M benchmark is a clear strength, providing a large-scale, aligned dataset for systematic evaluation of budget-constrained methods that future work can build upon.

major comments (2)

[Abstract] Abstract: the central performance-cost trade-off claim is asserted without any quantitative metrics, error bars, ablation results, or baseline comparisons, which prevents verification of whether the coupled formulation actually delivers the claimed advantage.
[Method] Method section (formulation): the unified coupling of sampling and cross-patch prediction is presented as addressing fragmented representations, yet the manuscript does not provide a concrete argument or test showing that LR-derived context plus the prediction head can recover task-critical intra-patch details (e.g., small objects or subtle boundaries) that remain ambiguous at low resolution; this assumption is load-bearing for the efficiency claim.

minor comments (1)

[Abstract] Abstract: the acronym 'GL' in GL-10M is never expanded, and the exact spatial alignment procedure or sensor sources for the 10 million images are not summarized.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We address the major comments point-by-point below and outline the revisions we plan to make.

read point-by-point responses

Referee: [Abstract] Abstract: the central performance-cost trade-off claim is asserted without any quantitative metrics, error bars, ablation results, or baseline comparisons, which prevents verification of whether the coupled formulation actually delivers the claimed advantage.

Authors: We agree that including quantitative evidence in the abstract would strengthen the presentation of our central claim. Although the abstract is intended to be concise, we will revise it to include specific performance-cost metrics from our experiments (e.g., accuracy improvements at given cost budgets and comparisons to baselines) to allow immediate verification of the advantage. The detailed results, including error bars and ablations, remain in the experimental section. revision: yes
Referee: [Method] Method section (formulation): the unified coupling of sampling and cross-patch prediction is presented as addressing fragmented representations, yet the manuscript does not provide a concrete argument or test showing that LR-derived context plus the prediction head can recover task-critical intra-patch details (e.g., small objects or subtle boundaries) that remain ambiguous at low resolution; this assumption is load-bearing for the efficiency claim.

Authors: This is a valid concern regarding the load-bearing assumption. Our experiments on recognition and retrieval tasks, which involve fine-grained details, demonstrate the effectiveness through superior trade-offs. However, to provide a more concrete argument, we will add in the revised manuscript a new analysis section with qualitative visualizations and quantitative metrics showing how the cross-patch prediction recovers intra-patch details that are ambiguous in LR, such as examples of small object detection and boundary precision under sparse HR sampling. revision: yes

Circularity Check

0 steps flagged

New problem formulation with independent experimental validation

full rationale

The paper presents a new unified cost-aware formulation for cross-scale remote sensing understanding that couples HR sampling decisions with cross-patch representation prediction. No equations, fitted parameters, or self-citations are invoked in the provided text to derive the core method; the approach is introduced as an original problem statement rather than a quantity computed from its own outputs. The GL-10M benchmark and recognition/retrieval experiments serve as external validation, keeping the derivation chain self-contained without reduction to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Central claim rests on the assumption that LR-derived global perception plus cross-patch prediction is sufficient to guide HR sampling; no free parameters, axioms, or invented entities are explicitly listed in the abstract.

invented entities (1)

GL-10M benchmark no independent evidence
purpose: Large-scale dataset of spatially aligned multi-resolution images for evaluating budget-constrained cross-scale reasoning
New dataset introduced to support systematic evaluation; no independent evidence of its construction or release is provided in the abstract.

pith-pipeline@v0.9.0 · 5508 in / 1206 out tokens · 61176 ms · 2026-05-10T15:13:24.776073+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

47 extracted references · 7 canonical work pages · 3 internal anchors

[1]

Arto Arampacha, Vidhani Dev, Goutham Venkatesh, Bhaskar Mayank, Ghosh Ritobrata, and Pal Sujit. 2021. Fine tuning clip with remote sensing (satellite) images and captions

2021
[2]

Mido Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zho- lus, et al. 2025. V-jepa 2: Self-supervised video models enable understanding, prediction and planning.arXiv preprint arXiv:2506.09985(2025)

work page internal anchor Pith review arXiv 2025
[3]

Mahmoud Assran, Quentin Duval, Ishan Misra, Piotr Bojanowski, Pascal Vincent, Michael Rabbat, Yann LeCun, and Nicolas Ballas. 2023. Self-supervised learning from images with a joint-embedding predictive architecture. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 15619–15629

2023
[4]

Adrien Bardes, Quentin Garrido, Jean Ponce, Xinlei Chen, Michael Rabbat, Yann LeCun, Mido Assran, and Nicolas Ballas. 2024. V-JEPA: latent video prediction for visual representation learning. InURL https://openreview. net/forum

2024
[5]

Favyen Bastani, Piper Wolters, Ritwik Gupta, Joe Ferdinando, and Aniruddha Kembhavi. 2023. Satlaspretrain: A large-scale dataset for remote sensing im- age understanding. InProceedings of the IEEE/CVF International Conference on Computer Vision. 16772–16782

2023
[6]

Davis Blalock, Jose Javier Gonzalez Ortiz, Jonathan Frankle, and John Guttag
[7]

What is the state of neural network pruning?Proceedings of machine learning and systems2 (2020), 129–146

2020
[8]

Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuhmann, Ludwig Schmidt, and Jenia Jitsev
[9]

In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Reproducible scaling laws for contrastive language-image learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2818–2829
[10]

Yezhen Cong, Samar Khanna, Chenlin Meng, Patrick Liu, Erik Rozi, Yutong He, Marshall Burke, David Lobell, and Stefano Ermon. 2022. Satmae: Pre-training transformers for temporal and multi-spectral satellite imagery.Advances in Neural Information Processing Systems35 (2022), 197–211

2022
[11]

Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020)

work page internal anchor Pith review Pith/arXiv arXiv 2020
[12]

Anthony Fuller, Koreen Millard, and James R Green. 2022. Satvit: Pretraining transformers for earth observation.IEEE Geoscience and Remote Sensing Letters 19 (2022), 1–5

2022
[13]

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition. 770–778

2016
[14]

Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. 2019. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing12, 7 (2019), 2217–2226

2019
[15]

Yuan Hu, Jianlong Yuan, Congcong Wen, Xiaonan Lu, Yu Liu, and Xiang Li. 2025. Rsgpt: A remote sensing vision language model and benchmark.ISPRS Journal of Photogrammetry and Remote Sensing224 (2025), 272–286

2025
[16]

Ziyue Huang, Hongxi Yan, Qiqi Zhan, Shuai Yang, Mingming Zhang, Chenkai Zhang, YiMing Lei, Zeming Liu, Qingjie Liu, and Yunhong Wang. 2025. A survey on remote sensing foundation models: From vision to multimodality.arXiv preprint arXiv:2503.22081(2025)

work page arXiv 2025
[17]

Jinseong Jang, Chunfei Ma, and Byeongwon Lee. 2025. Vl2lite: Task-specific knowledge distillation from large vision-language models to lightweight net- works. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 30073–30083

2025
[18]

Zhong Ji, Changxu Meng, Yan Zhang, Haoran Wang, Yanwei Pang, and Jun- gong Han. 2024. Eliminate before align: A remote sensing image-text retrieval framework with keyword explicit reasoning. InProceedings of the 32nd ACM international conference on multimedia. 1662–1671

2024
[19]

Zhenglun Kong, Peiyan Dong, Xiaolong Ma, Xin Meng, Wei Niu, Mengshu Sun, Xuan Shen, Geng Yuan, Bin Ren, Hao Tang, et al. 2022. Spvit: Enabling faster vision transformers via latency-aware soft token pruning. InEuropean conference on computer vision. Springer, 620–640

2022
[20]

Junyan Li, Delin Chen, Tianle Cai, Peihao Chen, Yining Hong, Zhenfang Chen, Yikang Shen, and Chuang Gan. 2024. Flexattention for efficient high-resolution vision-language models. InEuropean Conference on Computer Vision. Springer, 286–302

2024
[21]

Yansheng Li, Junwei Luo, Yongjun Zhang, Yihua Tan, Jin-Gang Yu, and Song Bai
[22]

Learning to holistically detect bridges from large-size VHR remote sensing imagery.IEEE transactions on pattern analysis and machine intelligence46, 12 (2024), 11507–11523

2024
[23]

Fan Liu, Delong Chen, Zhangqingyun Guan, Xiaocong Zhou, Jiale Zhu, Qiaolin Ye, Liyong Fu, and Jun Zhou. 2024. Remoteclip: A vision language foundation model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing 62 (2024), 1–16

2024
[24]

Ruixun Liu, Bowen Fu, Jiayi Song, Kaiyu Li, Wanchen Li, Lanxuan Xue, Hui Qiao, Weizhan Zhang, Deyu Meng, and Xiangyong Cao. 2025. ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks.arXiv preprint arXiv:2511.12267(2025)

work page arXiv 2025
[25]

Yinhe Liu, Yanfei Zhong, Sunan Shi, and Liangpei Zhang. 2024. Scale-aware deep reinforcement learning for high resolution remote sensing imagery classification. ISPRS Journal of Photogrammetry and Remote Sensing209 (2024), 296–311

2024
[26]

Zhunga Liu, Kun Li, Longfei Wang, and Zuowei Zhang. 2024. Multi-scale align- ment domain adaptation for ship classification in multi-resolution SAR images. IEEE Transactions on Automation Science and Engineering22 (2024), 4051–4062

2024
[27]

Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF international conference on computer vision. 10012–10022

2021
[28]

Utkarsh Mall, Bharath Hariharan, and Kavita Bala. 2023. Change-aware sampling and contrastive learning for satellite images. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 5261–5270

2023
[29]

Utkarsh Mall, Cheng Perng Phoo, Meilin Kelsey Liu, Carl Vondrick, Bharath Hariharan, and Kavita Bala. 2023. Remote sensing vision-language founda- tion models without annotations via ground remote alignment.arXiv preprint arXiv:2312.06960(2023)

work page arXiv 2023
[30]

Oscar Manas, Alexandre Lacoste, Xavier Giró-i Nieto, David Vazquez, and Pau Rodriguez. 2021. Seasonal contrast: Unsupervised pre-training from uncurated remote sensing data. InProceedings of the IEEE/CVF international conference on computer vision. 9414–9423

2021
[31]

Chenlin Meng, Enci Liu, Willie Neiswanger, Jiaming Song, Marshall Burke, David Lobell, and Stefano Ermon. 2022. IS-Count: Large-Scale Object Counting from Satellite Images with Covariate-Based Importance Sampling. (2022)

2022
[32]

Marco Mistretta, Alberto Baldrati, Marco Bertini, and Andrew D Bagdanov
[33]

InEuropean conference on computer vision

Improving zero-shot generalization of learned prompts via unsupervised knowledge distillation. InEuropean conference on computer vision. Springer, 459– 477
[34]

OpenStreetMap contributors. 2024. Planet dump retrieved from https://planet.openstreetmap.org. https://www.openstreetmap.org Data accessed 2024

2024
[35]

Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El- Nouby, et al. 2023. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193(2023)

work page internal anchor Pith review Pith/arXiv arXiv 2023
[36]

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sand- hini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al
[37]

In International conference on machine learning

Learning transferable visual models from natural language supervision. In International conference on machine learning. PmLR, 8748–8763
[38]

Colorado J Reed, Ritwik Gupta, Shufan Li, Sarah Brockman, Christopher Funk, Brian Clipp, Kurt Keutzer, Salvatore Candido, Matt Uyttendaele, and Trevor Darrell. 2023. Scale-mae: A scale-aware masked autoencoder for multiscale geospatial representation learning. InProceedings of the IEEE/CVF International Conference on Computer Vision. 4088–4099

2023
[39]

Shreelekha Revankar, Cheng Perng Phoo, Utkarsh Mall, Bharath Hariharan, and Kavita Bala. 2025. Scale-aware recognition in satellite images under resource constraints. In13th International Conference on Learning Representations, ICLR

2025
[40]

International Conference on Learning Representations, ICLR, 71438–71450
[41]

Ozan Sener and Silvio Savarese. 2018. Active Learning for Convolutional Neu- ral Networks: A Core-Set Approach. InInternational Conference on Learning Zhenghao Xie, Jing Xiao, Zhenqi Wang, Kexin Ma, Liang Liao, Gui-Song Xia, and Mi Wang Representations

2018
[42]

Gencer Sumbul, Marcela Charfuelan, Begüm Demir, and Volker Markl. 2019. Bigearthnet: A large-scale benchmark archive for remote sensing image under- standing.arXiv preprint arXiv:1902.06148(2019)

work page arXiv 2019
[43]

Xian Sun, Peijin Wang, Wanxuan Lu, Zicong Zhu, Xiaonan Lu, Qibin He, Junxi Li, Xuee Rong, Zhujun Yang, Hao Chang, et al. 2022. RingMo: A remote sensing foundation model with masked image modeling.IEEE Transactions on Geoscience and Remote Sensing61 (2022), 1–22

2022
[44]

Maofeng Tang, Andrei Cozma, Konstantinos Georgiou, and Hairong Qi. 2023. Cross-scale mae: A tale of multiscale exploitation in remote sensing.Advances in Neural Information Processing Systems36 (2023), 20054–20066

2023
[45]

Burak Uzkent and Stefano Ermon. 2020. Learning when and where to zoom with deep reinforcement learning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 12345–12354

2020
[46]

Anrui Wang, Yang Xu, He Wang, Zebin Wu, and Zhihui Wei. 2024. CDE-DETR: A real-time end-to-end high-resolution remote sensing object detection method based on RT-DETR. InIGARSS 2024-2024 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 8090–8094

2024
[47]

Zilun Zhang, Tiancheng Zhao, Yulong Guo, and Jianwei Yin. 2024. RS5M and GeoRSCLIP: A large-scale vision-language dataset and a large vision-language model for remote sensing.IEEE Transactions on Geoscience and Remote Sensing 62 (2024), 1–23. Observe Less, Understand More: Cost-aware Cross-scale Observation for Remote Sensing Understanding Appendix This a...

2024