Pith. sign in

REVIEW 4 major objections 5 minor 101 references

Satellites Reveal Mobility: A Commuting Origin-destination Flow Generator for Global Cities

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Public satellite imagery plus population counts can generate commuting origin-destination flows with over 98% of the expressiveness of traditional hard-to-collect urban data, via the GlODGen pipeline.

desk verdict The paper's core claim—that satellite imagery, not just population, carries the mobility signal—is plausible but never actually tested, because the experiments omit a population-only WEDAN baseline and the transfer comparison gives baselines only population and distance. read the letter →

arxiv 2505.15870 v1 pith:CHBKTWRE submitted 2025-05-21 cs.CV cs.CYeess.IV

Can Rong , Xin Zhang , Yanxin Xi , Hongjie Sui , Jingtao Ding , Yong Li This is my paper
classification cs.CVcs.CYeess.IV
keywords commutingorigin-destinationflowssatelliteimageryurbansemanticfeaturesvision-languagegeo-foundationmodelsgraphdiffusioncross-continentalgeneralizationpopulationdataglobalcities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Commuting origin-destination (OD) flows—how many people travel from each home region to each workplace—are essential for urban planning, traffic management, and sustainability research, but they are usually obtained through expensive travel surveys or privacy-sensitive mobile phone records. The paper claims that publicly available satellite imagery, combined with population counts, can stand in for the costly package of sociodemographic, economic, land-use, and point-of-interest data normally used to generate OD flows, recovering over 98% of that data's expressiveness in the best-performing model. To make the claim practical, the authors build GlODGen, a pipeline that crops satellite images to region boundaries, extracts urban semantic features with a pre-trained remote-sensing vision-language encoder, and generates OD flows with a graph diffusion model. Experiments in the U.S., U.K., China, Brazil, and Senegal support the claim that the pipeline transfers across continents and produces flows that track observed commuting patterns. If this holds, any city with publicly visible boundaries could obtain a commuting flow matrix without a single survey.

What carries the argument

The machinery that carries the argument is the pairing of RemoteCLIP, a pre-trained vision-language model for remote sensing whose frozen vision encoder turns each masked satellite image into a 1024-dimensional semantic vector, with WEDAN, a graph denoising diffusion model that generates OD flows as weighted directed edges of a city graph. Region representations are formed by concatenating the semantic vector with population; these representations condition the diffusion denoising process, in which a graph transformer predicts the noise to remove at each step. The mask-and-crop preprocessing step matters because it removes off-region pixels so the encoder sees only the region's own urban form. The roles are complementary: RemoteCLIP supplies general visual semantics without any mobility-specific training, while the diffusion model supplies the spatial structure of commuting flows.

What would settle it

Run GlODGen on the U.S. test split of the paper's benchmark twice, once with the true RemoteCLIP features and once with those features replaced by random vectors of the same dimension while population is kept unchanged. If the common part of commuting (CPC) metric does not drop by a statistically significant amount in the random-feature run, the claim that satellite imagery carries the mobility signal is falsified.

Watch

Extended reading notes

Core claim

The central discovery is that commuting OD flows can be generated from two globally available inputs—satellite imagery and population counts—almost as accurately as from the traditional bundle of sociodemographics, economics, land use, and POIs. Across data-driven OD generation models, the public inputs recover more than 90% of the performance of the traditional inputs, and with the best model (WEDAN) they reach 98.3% of its CPC score. The paper interprets this as evidence that satellite imagery contains rich urban semantic signals—residential versus commercial structure, density, built-up patterns—that encode where people live and work. GlODGen makes this operational by cropping satellite tiles to region boundaries, encoding each region with a frozen remote-sensing vision-language encoder, concatenating population, and letting a graph denoising diffusion model generate the directed weighted OD graph. The claim is that this transfers across continents and to cities with very different data environments.

Load-bearing premise

The load-bearing premise is that public satellite imagery, as read by a generic pre-trained remote-sensing encoder, carries enough mobility-relevant urban signal in every city that hard-to-collect survey, socioeconomic, land-use, and point-of-interest data are unnecessary; if that visual signal is absent or unreadable in some urban context, the generator has no information to fall back on.

Editorial extensions

If this is right

  • Cities with no travel surveys and no POI, land-use, or socioeconomic databases can obtain commuting OD matrices from region boundaries, public satellite imagery, and population counts alone.
  • A model trained on one continent can be transferred to another: in the U.S.-to-U.K. experiment GlODGen improved CPC by 34.0% and RMSE/NRMSE by 28.5% over baselines that only see population and distance.
  • Data-driven OD generation no longer needs to be confined to data-rich cities, since globally available inputs preserve most of the accuracy of expensive feature sets.
  • The generator produces static commuting flows only, so dynamic, time-dependent mobility questions remain outside its current scope.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same visual-semantic signal could plausibly support other flow tasks—freight, migration, epidemic spread—because the image encoder is frozen and not mobility-specific; this is an extension the authors do not test.
  • The 98% expressiveness result comes from U.S. data on which the generation model was also trained; the case-study correlations vary from 0.361 (London) to 0.816 (Rio), so the published evidence leaves open how much of the transfer success is visual signal and how much is the U.S.-shaped training distribution.
  • A natural testable extension is to train on multiple continents and see whether the low-correlation cities improve; if they do not, the bottleneck is likely the reference data or the visual signal, not the diffusion model.
  • Because the image encoder is frozen, GlODGen's output quality should improve automatically as better geo-foundation models become available, making these results a baseline for the architecture rather than a ceiling.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces GlODGen, a pipeline that replaces traditional hard-to-collect urban features (sociodemographics, economics, land use, POIs) with publicly available satellite imagery and population data for commuting origin-destination (OD) flow generation. Region-level satellite images are processed by the frozen RemoteCLIP vision encoder to produce 1024-dimension semantic features, which are concatenated with population and fed into WEDAN, a graph denoising diffusion model. Experiments compare models with public versus traditional features on US data (Table 1), test US-to-UK transfer (Table 2), and present rank-correlation case studies for Beijing, Shanghai, London, Paris, Rio, and Senegal (Figure 4). The paper claims that public data achieve over 98% expressiveness of traditional multisource data and that GlODGen generalizes across continents; the code is released.

Significance. If the central claim is fully supported, the work would substantially lower the data-access barrier for commuting OD flow generation, enabling city-scale flow matrices for data-scarce regions and supporting applications in urban planning, resilience, and sustainability research. The open-source release, the use of frozen foundation-model features, and the multi-continent evaluation are strengths. However, the significance is currently tempered by the absence of an ablation that isolates the satellite-imagery contribution from population and architecture capacity, and by an uncontrolled cross-continental comparison that conflates input modality with model family. The contribution is empirical rather than methodological, since the components (RemoteCLIP and WEDAN) are off-the-shelf.

major comments (4)
  1. [Section 4.1, Table 1] The public-data condition always consists of RemoteCLIP features plus population, and the traditional condition also includes population; no model is run with population alone. Consequently, the 98.3% CPC ratio for WEDAN cannot be attributed to the satellite imagery; it may reflect the WEDAN architecture or population information. The authors should add a population-only (and ideally population-plus-distance) condition for WEDAN and other models to isolate the contribution of the image features, and should temper the Abstract's 'over 98%' claim accordingly.
  2. [Section 4.2, Table 2] The cross-continental comparison is not controlled: GlODGen receives satellite imagery and population and is built on WEDAN, whereas the baselines receive only population and inter-region distances and use different model families. The reported 46.5% CPC improvement over GMEL therefore conflates input features with model capacity. A fair test requires comparing GlODGen against WEDAN trained on the same restricted inputs (population and distance) and ideally also against a version of GlODGen without satellite features.
  3. [Section 4.3, Figure 4] The case-study correlations are rank-based summaries without error bars or significance testing, and values such as London (0.361) and Paris (0.465) are modest. The text's assertion that 'high Spearman rank correlations observed in most cases' is not fully supported by the figure. The authors should report confidence intervals or null-model comparisons, and should avoid over-interpreting the visual resemblance given the acknowledged heterogeneity of reference data.
  4. [Abstract and Section 4.1] The claim of 'over 98% expressiveness' is based on WEDAN's 98.3% CPC in Table 1; the other data-driven models range from 90.5% to 97.6%. The abstract cherry-picks the best-performing model. The paper should state the range or explicitly attribute the value to WEDAN and update the narrative to reflect that the gap is small but not uniformly below 2%.
minor comments (5)
  1. [Section 5] The phrase 'Boarder Impact' should read 'Broader Impact'.
  2. [Section 2.1, Definition 3] The flow notation 'Frorg ,rdst' contains formatting issues and should be typeset cleanly.
  3. [Table 2 caption] The caption's phrase 'geographically distributed population' is unclear; the caption should state explicitly that baselines receive only population and distance.
  4. [Appendix B.4] The disclosure that some visualizations also appear in a separate submission under review should be resolved or clearly delineated before publication to avoid ambiguity about overlapping contributions.
  5. [Figure 4] The smoothing procedure for the rank-aligned normalized flow curves is not described; please provide details in the text or caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical comparison on held-out and cross-continental data, not a fitted quantity or self-citation chain.

full rationale

The paper's derivation chain is empirical and self-contained. The load-bearing claim — that public satellite imagery plus population can replace costly sociodemographic, economic, land-use, and POI inputs — is tested by comparing WEDAN trained on RemoteCLIP-derived public features against WEDAN trained on traditional ACS/POI features on held-out US test urban areas (Table 1), and by US-to-UK cross-continental transfer against population/distance baselines (Table 2). No predictive quantity is fitted to the target it is then said to predict: RemoteCLIP weights are frozen, population comes from census data, and generated OD flows are evaluated against LODES, ONS, INSEE, and CDR references that are not used for training. Self-citations ([55], [56], [57]) are used to import a published diffusion model and training conventions; they are not invoked as uniqueness theorems and do not forbid alternative architectures. The absence of a population-only WEDAN ablation is an experimental limitation — the satellite modality's marginal contribution is not separately isolated — but a missing control is not a definitional equivalence or a fitted-input renaming. No circular step can be exhibited from the paper's own equations or citations.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several empirical and domain assumptions about the representational power of satellite imagery and the transferability of the US-trained model. No new theoretical entities or fitted constants are introduced; the system is an empirical pipeline.

assumptions (4)
  • domain assumption Satellite imagery encodes urban semantics that are predictive of commuting flows.
    Invoked in Section 1 and throughout; the entire pipeline depends on this, but it is never formally established beyond the empirical evaluations.
  • domain assumption The vision-language geo-foundation model RemoteCLIP produces features that are transferable across cities and continents.
    The cross-continental transfer experiments in Section 4.2 rely on this; RemoteCLIP is used with frozen weights.
  • domain assumption The US dataset (LODES) is representative enough to train a globally applicable generator.
    The model is trained solely on US data, yet the goal is global applicability (Section 4.2).
  • domain assumption Commuting OD flows are static and stable over time.
    Definition 3 states flows 'remain relatively stable for a long time'; this underlies the generation target.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Satellites Reveal Mobility: A Commuting Origin-destination Flow Generator for Global Cities." pith.science (2026). https://pith.science/paper/CHBKTWRE

@misc{pith2026250515870,
  author       = {Pith},
  title        = {Pith review of: Satellites Reveal Mobility: A Commuting Origin-destination Flow Generator for Global Cities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CHBKTWRE}},
  note         = {Machine review of arXiv:2505.15870}
}
read the original abstract

Commuting Origin-destination~(OD) flows, capturing daily population mobility of citizens, are vital for sustainable development across cities around the world. However, it is challenging to obtain the data due to the high cost of travel surveys and privacy concerns. Surprisingly, we find that satellite imagery, publicly available across the globe, contains rich urban semantic signals to support high-quality OD flow generation, with over 98\% expressiveness of traditional multisource hard-to-collect urban sociodemographic, economics, land use, and point of interest data. This inspires us to design a novel data generator, GlODGen, which can generate OD flow data for any cities of interest around the world. Specifically, GlODGen first leverages Vision-Language Geo-Foundation Models to extract urban semantic signals related to human mobility from satellite imagery. These features are then combined with population data to form region-level representations, which are used to generate OD flows via graph diffusion models. Extensive experiments on 4 continents and 6 representative cities show that GlODGen has great generalizability across diverse urban environments on different continents and can generate OD flow data for global cities highly consistent with real-world mobility data. We implement GlODGen as an automated tool, seamlessly integrating data acquisition and curation, urban semantic feature extraction, and OD flow generation together. It has been released at https://github.com/tsinghua-fib-lab/generate-od-pubtools.

Figures

Figures reproduced from arXiv: 2505.15870 by the authors.

Figure 1
Figure 1. An illustration of replacing the traditional hard-to-collect urban data by easily accessible [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Cases of satellite imagery for urban regions. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The framework and data pipeline of GlODGen. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Correlation analysis between generated OD flows and the flows extracted from diverse [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the generated and data-oriented OD flows in Beijing and Shanghai. The [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the generated and data-oriented OD flows in London and Paris. The [PITH_FULL_IMAGE:figures/full_fig_p026_6.png]
Figure 7
Figure 7. Figure 7: Visualization of the generated and data-oriented OD flows in Rio de Janeiro. The generated [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the generated and data-oriented OD flows in Senegal. The generated OD [PITH_FULL_IMAGE:figures/full_fig_p026_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

101 extracted references · 56 canonical work pages

  1. [1]

    Interpretable socioeconomic status inference from aerial imagery through urban patterns

    Jacob Levy Abitbol and Marton Karsai. Interpretable socioeconomic status inference from aerial imagery through urban patterns. Nature Machine Intelligence, 2(11):684–692, 2020

    2020

  2. [2]

    A human-machine collaborative approach measures economic development using satellite imagery

    Donghyun Ahn, Jeasurk Yang, Meeyoung Cha, Hyunjoo Yang, Jihee Kim, Sangyoon Park, Sungwon Han, Eunji Lee, Susang Lee, and Sungwon Park. A human-machine collaborative approach measures economic development using satellite imagery. Nature Communications, 14(1):6811, 2023

    2023

  3. [3]

    Observing the rhythms of daily life: A six-week travel diary

    Kay W Axhausen, Andrea Zimmermann, Stefan Schönfelder, Guido Rindsfüser, and Thomas Haupt. Observing the rhythms of daily life: A six-week travel diary. Transportation, 29(2):95– 124, 2002

    2002

  4. [4]

    Gener- ating interpretable poverty maps using object detection in satellite images

    Kumar Ayush, Burak Uzkent, Marshall Burke, David Lobell, and Stefano Ermon. Gener- ating interpretable poverty maps using object detection in satellite images. arXiv preprint arXiv:2002.01612, 2020

    arXiv 2002

  5. [5]

    Efficient poverty mapping from high resolution remote sensing images

    Kumar Ayush, Burak Uzkent, Kumar Tanmay, Marshall Burke, David Lobell, and Stefano Ermon. Efficient poverty mapping from high resolution remote sensing images. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 12–20, 2021

    2021

  6. [6]

    Human mobility: Models and applications

    Hugo Barbosa, Marc Barthelemy, Gourab Ghoshal, Charlotte R James, Maxime Lenormand, Thomas Louail, Ronaldo Menezes, José J Ramasco, Filippo Simini, and Marcello Tomasini. Human mobility: Models and applications. Physics Reports, 734:1–74, 2018

    2018

  7. [7]

    Ef- ficient and scalable graph generation through iterative local expansion

    Andreas Bergmeister, Karolis Martinkus, Nathanaël Perraudin, and Roger Wattenhofer. Ef- ficient and scalable graph generation through iterative local expansion. arXiv preprint arXiv:2312.11529, 2023

    arXiv 2023

  8. [8]

    Netgan: Generating graphs via random walks

    Aleksandar Bojchevski, Oleksandr Shchur, Daniel Zügner, and Stephan Günnemann. Netgan: Generating graphs via random walks. In International conference on machine learning, pages 610–619. PMLR, 2018

    2018

Show all 101 references
  1. [9]

    Census Bureau

    U.S. Census Bureau. Lehd origin-destination employment statistics data (2002-2021), 2024

    2002

  2. [10]

    Using satellite imagery to understand and promote sustainable development

    Marshall Burke, Anne Driscoll, David B Lobell, and Stefano Ermon. Using satellite imagery to understand and promote sustainable development. Science, 371(6535):eabe8628, 2021

    2021

  3. [11]

    Estimating origin- destination flows using opportunistically collected mobile phone location data from one million users in boston metropolitan area

    Francesco Calabrese, Giusy Di Lorenzo, Liang Liu, and Carlo Ratti. Estimating origin- destination flows using opportunistically collected mobile phone location data from one million users in boston metropolitan area. 2011

    2011

  4. [12]

    Dynamic vulnerability analysis of public transport networks: mitigation effects of real-time information

    Oded Cats and Erik Jenelius. Dynamic vulnerability analysis of public transport networks: mitigation effects of real-time information. Networks and Spatial Economics , 14:435–463, 2014

    2014

  5. [13]

    Human mobility and socioeconomic datasets of the rio de janeiro metropolitan area

    Júlio César Chaves, Moacyr AHB da Silva, Ricardo de Souza Alencar, Alexandre G Evsukoff, and Vinícius da Fonseca Vieira. Human mobility and socioeconomic datasets of the rio de janeiro metropolitan area. Data in brief, 51:109695, 2023

    2023

  6. [14]

    Practical bottom-up golf course detection using multispectral remote sensing imagery

    Jingbo Chen, Chengyi Wang, Dongxu He, Jiansheng Chen, and Anzhi Yue. Practical bottom-up golf course detection using multispectral remote sensing imagery. In Image and Graphics: 9th International Conference, ICIG 2017, Shanghai, China, September 13-15, 2017, Revised Selected P...

    2017

  7. [15]

    Characterizing the temporally stable structure of community evolution in intra-urban origin- destination networks

    Xiao-Jian Chen, Yuhui Zhao, Chaogui Kang, Xiaoyue Xing, Quanhua Dong, and Yu Liu. Characterizing the temporally stable structure of community evolution in intra-urban origin- destination networks. Cities, 150:105033, 2024

    2024

  8. [16]

    D4d-senegal: the second mobile phone data for development challenge

    Yves-Alexandre de Montjoye, Zbigniew Smoreda, Romain Trinquart, Cezary Ziemlicki, and Vincent D Blondel. D4d-senegal: the second mobile phone data for development challenge. arXiv preprint arXiv:1407.4885, 2014. 10

    2014 arXiv

  9. [17]

    Don’t stop learning: Towards continual learning for the clip model

    Yuxuan Ding, Lingqiao Liu, Chunna Tian, Jingyuan Yang, and Haoxuan Ding. Don’t stop learning: Towards continual learning for the clip model. arXiv preprint arXiv:2207.09248, 2022

    2022 arXiv

  10. [18]

    An image is worth 16x16 words: Transformers for image recognition at scale

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

    2010 arXiv

  11. [19]

    A generalization of transformer networks to graphs

    Vijay Prakash Dwivedi and Xavier Bresson. A generalization of transformer networks to graphs. arXiv preprint arXiv:2012.09699, 2020

    2012 arXiv

  12. [20]

    Citybench: Evaluating the capabilities of large language model as world model

    Jie Feng, Jun Zhang, Junbo Yan, Xin Zhang, Tianjian Ouyang, Tianhui Liu, Yuwei Du, Siqi Guo, and Yong Li. Citybench: Evaluating the capabilities of large language model as world model. arXiv preprint arXiv:2406.13945, 2024

    2024 arXiv

  13. [21]

    Hyperbolic geometric latent diffusion model for graph generation

    Xingcheng Fu, Yisen Gao, Yuecen Wei, Qingyun Sun, Hao Peng, Jianxin Li, and Xianxian Li. Hyperbolic geometric latent diffusion model for graph generation. In Proceedings of the 41st International Conference on Machine Learning, ICML’24. JMLR.org, 2024

    2024

  14. [22]

    Travel demand estimation and network assignment based on cellular network data

    David Gundlegård, Clas Rydergren, Nils Breyer, and Botond Rajna. Travel demand estimation and network assignment based on cellular network data. Computer Communications, 95:29–42, 2016

    2016

  15. [23]

    Diffusion models for graphs benefit from discrete state spaces.arXiv preprint arXiv:2210.01549, 2022

    Kilian Konstantin Haefeli, Karolis Martinkus, Nathanaël Perraudin, and Roger Wattenhofer. Diffusion models for graphs benefit from discrete state spaces.arXiv preprint arXiv:2210.01549, 2022

    2022 arXiv

  16. [24]

    Lightweight and robust representation of economic scales from satellite imagery

    Sungwon Han, Donghyun Ahn, Hyunji Cha, Jeasurk Yang, Sungwon Park, and Meeyoung Cha. Lightweight and robust representation of economic scales from satellite imagery. InProceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 428–436, 2020

    2020

  17. [25]

    Learning to score economic development from satellite imagery

    Sungwon Han, Donghyun Ahn, Sungwon Park, Jeasurk Yang, Susang Lee, Jihee Kim, Hyunjoo Yang, Sangyoon Park, and Meeyoung Cha. Learning to score economic development from satellite imagery. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & D...

    2020

  18. [26]

    Urbanvlp: A multi-granularity vision-language pre-trained foundation model for urban indicator prediction

    Xixuan Hao, Wei Chen, Yibo Yan, Siru Zhong, Kun Wang, Qingsong Wen, and Yuxuan Liang. Urbanvlp: A multi-granularity vision-language pre-trained foundation model for urban indicator prediction. arXiv preprint arXiv:2403.16831, 2024

    2024 arXiv

  19. [27]

    Urbanvlp: Multi-granularity vision-language pretraining for urban socioeconomic indicator prediction

    Xixuan Hao, Wei Chen, Yibo Yan, Siru Zhong, Kun Wang, Qingsong Wen, and Yuxuan Liang. Urbanvlp: Multi-granularity vision-language pretraining for urban socioeconomic indicator prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 28061–...

    2025

  20. [28]

    Human mobility data and analysis for urban resilience: A systematic review

    Masahiko Haraguchi, Akihiko Nishino, Akira Kodaka, Maura Allaire, Upmanu Lall, Liao Kuei- Hsien, Kaya Onda, Kota Tsubouchi, and Naohiko Kohtake. Human mobility data and analysis for urban resilience: A systematic review. Environment and Planning B: Urban Analytics and City Sci...

    2022

  21. [29]

    Deep residual learning for image recognition

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

    2016

  22. [30]

    Perceiving commerial activeness over satellite images

    Zhiyuan He, Su Yang, Weishan Zhang, and Jiulong Zhang. Perceiving commerial activeness over satellite images. In Companion Proceedings of the The Web Conference 2018 , pages 387–394, 2018

    2018

  23. [31]

    Graph convolutional networks for hyperspectral image classification

    Danfeng Hong, Lianru Gao, Jing Yao, Bing Zhang, Antonio Plaza, and Jocelyn Chanussot. Graph convolutional networks for hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing, 59(7):5966–5978, 2020. 11

    2020

  24. [32]

    3d-llm: Injecting the 3d world into large language models

    Yining Hong, Haoyu Zhen, Peihao Chen, Shuhong Zheng, Yilun Du, Zhenfang Chen, and Chuang Gan. 3d-llm: Injecting the 3d world into large language models. Advances in Neural Information Processing Systems, 36:20482–20494, 2023

    2023

  25. [33]

    Open synthetic travel demand for paris and île-de-france: Inputs and output data

    Sebastian Hörl and Milos Balac. Open synthetic travel demand for paris and île-de-france: Inputs and output data. Data in Brief, 39:107622, 2021

    2021

  26. [34]

    Development of origin–destination matrices using mobile phone call data

    Md Shahadat Iqbal, Charisma F Choudhury, Pu Wang, and Marta C González. Development of origin–destination matrices using mobile phone call data. Transportation Research Part C: Emerging Technologies, 40:63–74, 2014

    2014

  27. [35]

    Conceptfusion: Open-set multimodal 3d mapping

    Krishna Murthy Jatavallabhula, Alihusein Kuwajerwala, Qiao Gu, Mohd Omama, Tao Chen, Alaa Maalouf, Shuang Li, Ganesh Iyer, Soroush Saryazdi, Nikhil Keetha, et al. Conceptfusion: Open-set multimodal 3d mapping. arXiv preprint arXiv:2302.07241, 2023

    2023 arXiv

  28. [36]

    Combining satellite imagery and machine learning to predict poverty

    Neal Jean, Marshall Burke, Michael Xie, W Matthew Davis, David B Lobell, and Stefano Ermon. Combining satellite imagery and machine learning to predict poverty. Science, 353(6301):790– 794, 2016

    2016

  29. [37]

    Population flow drives spatio-temporal distribution of covid-19 in china

    Jayson S Jia, Xin Lu, Yun Yuan, Ge Xu, Jianmin Jia, and Nicholas A Christakis. Population flow drives spatio-temporal distribution of covid-19 in china. Nature, 582(7812):389–394, 2020

    2020

  30. [38]

    Score-based generative modeling of graphs via the system of stochastic differential equations

    Jaehyeong Jo, Seul Lee, and Sung Ju Hwang. Score-based generative modeling of graphs via the system of stochastic differential equations. In International conference on machine learning, pages 10362–10383. PMLR, 2022

    2022

  31. [39]

    Object detection in optical remote sensing images: A survey and a new benchmark

    Ke Li, Gang Wan, Gong Cheng, Liqiu Meng, and Junwei Han. Object detection in optical remote sensing images: A survey and a new benchmark. ISPRS journal of photogrammetry and remote sensing, 159:296–307, 2020

    2020

  32. [40]

    Learning representations of satellite imagery by leveraging point-of-interests

    Tong Li, Yanxin Xi, Huandong Wang, Yong Li, Sasu Tarkoma, and Pan Hui. Learning representations of satellite imagery by leveraging point-of-interests. ACM Transactions on Intelligent Systems and Technology, 14(4):1–32, 2023

    2023

  33. [41]

    Predicting multi-level socioeconomic indicators from structural urban imagery

    Tong Li, Shiduo Xin, Yanxin Xi, Sasu Tarkoma, Pan Hui, and Yong Li. Predicting multi-level socioeconomic indicators from structural urban imagery. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management, pages 3282–3291, 2022

    2022

  34. [42]

    Vision-language models in remote sensing: Current progress and future trends

    Xiang Li, Congcong Wen, Yuan Hu, Zhenghang Yuan, and Xiao Xiang Zhu. Vision-language models in remote sensing: Current progress and future trends. IEEE Geoscience and Remote Sensing Magazine, 2024

    2024

  35. [43]

    Remoteclip: A vision language foundation model for remote sensing

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

    2024

  36. [44]

    Learning geo-contextual embeddings for commuting flow prediction

    Zhicheng Liu, Fabio Miranda, Weiting Xiong, Junyan Yang, Qiao Wang, and Claudio Silva. Learning geo-contextual embeddings for commuting flow prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, pages 808–816, 2020

    2020

  37. [45]

    Semantic segmentation of crop type in africa: A novel dataset and analysis of deep learning methods

    Rose M Rustowicz, Robin Cheong, Lijing Wang, Stefano Ermon, Marshall Burke, and David Lobell. Semantic segmentation of crop type in africa: A novel dataset and analysis of deep learning methods. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2019

  38. [46]

    Fully convolutional recurrent networks for mul- tidate crop recognition from multitemporal image sequences.ISPRS Journal of Photogrammetry and Remote Sensing, 171:188–201, 2021

    Jorge Andres Chamorro Martinez, Laura Elena Cué La Rosa, Raul Queiroz Feitosa, Ieda Del’Arco Sanches, and Patrick Nigri Happ. Fully convolutional recurrent networks for mul- tidate crop recognition from multitemporal image sequences.ISPRS Journal of Photogrammetry and Remote S...

    2021

  39. [47]

    Permutation invariant graph generation via score-based generative modeling

    Chenhao Niu, Yang Song, Jiaming Song, Shengjia Zhao, Aditya Grover, and Stefano Ermon. Permutation invariant graph generation via score-based generative modeling. In International conference on artificial intelligence and statistics, pages 4474–4484. PMLR, 2020. 12

    2020

  40. [48]

    User guide to origin-destination data for census 2021, england and wales

    Office for National Statistics. User guide to origin-destination data for census 2021, england and wales. https://www.ons.gov.uk, October 2023. Accessed: 2023-10-26

    2021

  41. [49]

    Cellular-based data-extracting method for trip distribution

    Changxuan Pan, Jiangang Lu, Shan Di, and Bin Ran. Cellular-based data-extracting method for trip distribution. Transportation research record, 1945(1):33–39, 2006

    1945

  42. [50]

    Poverty prediction with public landsat 7 satellite imagery and machine learning

    Anthony Perez, Christopher Yeh, George Azzari, Marshall Burke, David Lobell, and Stefano Ermon. Poverty prediction with public landsat 7 satellite imagery and machine learning. arXiv preprint arXiv:1711.03654, 2017

    2017 arXiv

  43. [51]

    Trip distribution modeling with twitter data.Computers, Environment and Urban Systems, 77:101354, 2019

    Nastaran Pourebrahim, Selima Sultana, Amirreza Niakanlahiji, and Jean-Claude Thill. Trip distribution modeling with twitter data.Computers, Environment and Urban Systems, 77:101354, 2019

    2019

  44. [52]

    Urban resilience: A conceptual framework

    Paulo Jorge Gomes Ribeiro and Luís António Pena Jardim Gonçalves. Urban resilience: A conceptual framework. Sustainable Cities and Society, 50:101625, 2019

    2019

  45. [53]

    A machine learning approach to modeling human migration

    Caleb Robinson and Bistra Dilkina. A machine learning approach to modeling human migration. In Proceedings of the 1st ACM SIGCAS Conference on Computing and Sustainable Societies, pages 1–8, 2018

    2018

  46. [54]

    A generalizable and accessible approach to machine learning with global satellite imagery

    Esther Rolf, Jonathan Proctor, Tamma Carleton, Ian Bolliger, Vaishaal Shankar, Miyabi Ishihara, Benjamin Recht, and Solomon Hsiang. A generalizable and accessible approach to machine learning with global satellite imagery. Nature communications, 12(1):4392, 2021

    2021

  47. [55]

    An interdisciplinary survey on origin-destination flows modeling: Theory and techniques

    Can Rong, Jingtao Ding, and Yong Li. An interdisciplinary survey on origin-destination flows modeling: Theory and techniques. arXiv preprint arXiv:2306.10048, 2023

    2023 arXiv

  48. [56]

    A large-scale dataset and benchmark for commuting origin-destination flow generation

    Can Rong, Jingtao Ding, Yan Liu, and Yong Li. A large-scale dataset and benchmark for commuting origin-destination flow generation. In The Thirteenth International Conference on Learning Representations

  49. [57]

    City-wide origin-destination matrix generation via graph denoising diffusion

    Can Rong, Jingtao Ding, Zhicheng Liu, and Yong Li. City-wide origin-destination matrix generation via graph denoising diffusion. arXiv preprint arXiv:2306.04873, 2023

    2023 arXiv

  50. [58]

    Goddag: Generating origin-destination flow for new cities via domain adversarial training

    Can Rong, Jie Feng, and Jingtao Ding. Goddag: Generating origin-destination flow for new cities via domain adversarial training. IEEE Transactions on Knowledge and Data Engineering, 2023

    2023

  51. [59]

    Self-attention for raw optical satellite time series classifica- tion

    Marc Rußwurm and Marco Körner. Self-attention for raw optical satellite time series classifica- tion. ISPRS journal of photogrammetry and remote sensing, 169:421–435, 2020

    2020

  52. [60]

    One-shot transfer learning for population mapping

    Erzhuo Shao, Jie Feng, Yingheng Wang, Tong Xia, and Yong Li. One-shot transfer learning for population mapping. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 1588–1597, 2021

    2021

  53. [61]

    Kidsat: satellite imagery to map childhood poverty dataset and benchmark

    Makkunda Sharma, Fan Yang, Duy-Nhat V o, Esra Suel, Swapnil Mishra, Samir Bhatt, Oliver Fiala, William Rudgard, and Seth Flaxman. Kidsat: satellite imagery to map childhood poverty dataset and benchmark. arXiv preprint arXiv:2407.05986, 2024

    2024 arXiv

  54. [62]

    M3luc: Multi-modal model for urban land-use classification

    Li Sibo, Zhang Xin, Lin Yuming, and Li Yong. M3luc: Multi-modal model for urban land-use classification. In Proceedings of the 32nd ACM International Conference on Advances in Geographic Information Systems, pages 1–12, 2024

    2024

  55. [63]

    A deep gravity model for mobility flows generation

    Filippo Simini, Gianni Barlacchi, Massimilano Luca, and Luca Pappalardo. A deep gravity model for mobility flows generation. Nature communications, 12(1):1–13, 2021

    2021

  56. [64]

    A universal model for mobility and migration patterns

    Filippo Simini, Marta C González, Amos Maritan, and Albert-László Barabási. A universal model for mobility and migration patterns. Nature, 484(7392):96–100, 2012

    2012

  57. [65]

    Very deep convolutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 13

    2014 arXiv

  58. [66]

    Understanding metropolitan patterns of daily encounters

    Lijun Sun, Kay W Axhausen, Der-Horng Lee, and Xianfeng Huang. Understanding metropolitan patterns of daily encounters. Proceedings of the National Academy of Sciences, 110(34):13774– 13779, 2013

    2013

  59. [67]

    Pbnet: Part-based convolutional neural network for complex composite object detection in remote sensing imagery

    Xian Sun, Peijin Wang, Cheng Wang, Yingfei Liu, and Kun Fu. Pbnet: Part-based convolutional neural network for complex composite object detection in remote sensing imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 173:50–65, 2021

    2021

  60. [68]

    Individus localisés au canton-ou-ville en 2015

    The National Institute of Statistics and Economic Studies (INSEE). Individus localisés au canton-ou-ville en 2015. https://www.insee.fr/fr/statistiques/3625223, 2015

    2015

  61. [69]

    Learning to interpret satellite images using wikipedia

    Burak Uzkent, Evan Sheehan, Chenlin Meng, Zhongyi Tang, Marshall Burke, David Lobell, and Stefano Ermon. Learning to interpret satellite images using wikipedia. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, 2019

    2019

  62. [70]

    Digress: Discrete denoising diffusion for graph generation

    Clement Vignac, Igor Krawczuk, Antoine Siraudin, Bohan Wang, V olkan Cevher, and Pas- cal Frossard. Digress: Discrete denoising diffusion for graph generation. arXiv preprint arXiv:2209.14734, 2022

    2022 arXiv

  63. [71]

    Deep transfer learning for crop yield prediction with remote sensing data

    Anna X Wang, Caelin Tran, Nikhil Desai, David Lobell, and Stefano Ermon. Deep transfer learning for crop yield prediction with remote sensing data. In Proceedings of the 1st ACM SIGCAS Conference on Computing and Sustainable Societies, pages 1–5, 2018

    2018

  64. [72]

    Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests

    Yanxin Xi, Tong Li, Huandong Wang, Yong Li, Sasu Tarkoma, and Pan Hui. Beyond the first law of geography: Learning representations of satellite imagery by leveraging point-of-interests. In Proceedings of the ACM Web Conference 2022, pages 3308–3316, 2022

    2022

  65. [73]

    A satellite imagery dataset for long-term sustainable development in united states cities

    Yanxin Xi, Yu Liu, Tong Li, Jintao Ding, Yunke Zhang, Sasu Tarkoma, Yong Li, and Pan Hui. A satellite imagery dataset for long-term sustainable development in united states cities. Scientific data, 10(1):866, 2023

    2023

  66. [74]

    From pixels to progress: generating road network from satellite imagery for socioeconomic insights in impoverished areas

    Yanxin Xi, Yu Liu, Zhicheng Liu, Sasu Tarkoma, Pan Hui, and Yong Li. From pixels to progress: generating road network from satellite imagery for socioeconomic insights in impoverished areas. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intell...

    2024

  67. [75]

    Dota: A large-scale dataset for object detection in aerial images

    Gui-Song Xia, Xiang Bai, Jian Ding, Zhen Zhu, Serge Belongie, Jiebo Luo, Mihai Datcu, Marcello Pelillo, and Liangpei Zhang. Dota: A large-scale dataset for object detection in aerial images. In Proceedings of the IEEE conference on computer vision and pattern recognition, page...

    2018

  68. [76]

    Refound: Crafting a foundation model for urban region understanding upon language and visual foundations

    Congxi Xiao, Jingbo Zhou, Yixiong Xiao, Jizhou Huang, and Hui Xiong. Refound: Crafting a foundation model for urban region understanding upon language and visual foundations. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 3527–3538, 2024

    2024

  69. [77]

    Urbanclip: Learning text-enhanced urban region profiling with contrastive language-image pretraining from the web

    Yibo Yan, Haomin Wen, Siru Zhong, Wei Chen, Haodong Chen, Qingsong Wen, Roger Zimmermann, and Yuxuan Liang. Urbanclip: Learning text-enhanced urban region profiling with contrastive language-image pretraining from the web. In Proceedings of the ACM on Web Conference 2024, page...

    2024

  70. [78]

    Sustainbench: Benchmarks for monitoring the sustainable development goals with machine learning

    Christopher Yeh, Chenlin Meng, Sherrie Wang, Anne Driscoll, Erik Rozi, Patrick Liu, Ji- hyeon Lee, Marshall Burke, David B Lobell, and Stefano Ermon. Sustainbench: Benchmarks for monitoring the sustainable development goals with machine learning. arXiv preprint arXiv:2111.04724, 2021

    2021 arXiv

  71. [79]

    Using publicly available satellite imagery and deep learning to understand economic well-being in africa

    Christopher Yeh, Anthony Perez, Anne Driscoll, George Azzari, Zhongyi Tang, David Lobell, Stefano Ermon, and Marshall Burke. Using publicly available satellite imagery and deep learning to understand economic well-being in africa. Nature communications, 11(1):2583, 2020

    2020

  72. [80]

    Deep gaussian process for crop yield prediction based on remote sensing data

    Jiaxuan You, Xiaocheng Li, Melvin Low, David Lobell, and Stefano Ermon. Deep gaussian process for crop yield prediction based on remote sensing data. In Proceedings of the AAAI conference on artificial intelligence, volume 31, 2017. 14

    2017

  73. [81]

    Estimating on-road transportation carbon emissions from open data of road network and origin-destination flow data

    Jinwei Zeng, Yu Liu, Jingtao Ding, Jian Yuan, and Yong Li. Estimating on-road transportation carbon emissions from open data of road network and origin-destination flow data. In Pro- ceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 14539–14547, 2024

    2024

  74. [82]

    Uv-sam: Adapting segment anything model for urban village identification

    Xin Zhang, Yu Liu, Yuming Lin, Qingmin Liao, and Yong Li. Uv-sam: Adapting segment anything model for urban village identification. arXiv preprint arXiv:2401.08083, 2024

    2024 arXiv

  75. [83]

    Urbancross: enhancing satellite image-text retrieval with cross-domain adaptation

    Siru Zhong, Xixuan Hao, Yibo Yan, Ying Zhang, Yangqiu Song, and Yuxuan Liang. Urbancross: enhancing satellite image-text retrieval with cross-domain adaptation. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 6307–6315, 2024

    2024

  76. [84]

    Towards vision-language geo-foundation model: A survey

    Yue Zhou, Litong Feng, Yiping Ke, Xue Jiang, Junchi Yan, Xue Yang, and Wayne Zhang. Towards vision-language geo-foundation model: A survey. arXiv preprint arXiv:2406.09385, 2024

    2024

  77. [85]

    The p 1 p 2/d hypothesis: on the intercity movement of persons

    George Kingsley Zipf. The p 1 p 2/d hypothesis: on the intercity movement of persons. American sociological review, 11(6):677–686, 1946. 15 NeurIPS Paper Checklist

    1946

  78. [86]

    Guidelines: • The answer NA means that the abstract and introduction do not include the claims made in the paper

    Claims Question: Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? Answer: [Yes] Justification: See Abstract and Section 1. Guidelines: • The answer NA means that the abstract and introduction do not include the cl...

  79. [87]

    Limitations

    Limitations Question: Does the paper discuss the limitations of the work performed by the authors? Answer: [Yes] Justification: See Section 5. Guidelines: • The answer NA means that the paper has no limitation while the answer No means that the paper has limitations, but those...

  80. [88]

    Guidelines: • The answer NA means that the paper does not include theoretical results

    Theory assumptions and proofs Question: For each theoretical result, does the paper provide the full set of assumptions and a complete (and correct) proof? Answer: [NA] 16 Justification: Our work is not about theoretical results. Guidelines: • The answer NA means that the pape...

  81. [89]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental result reproducibility Question: Does the paper fully disclose all the information needed to reproduce the main ex- perimental results of the paper to the extent that it affects the main claims and/or conclusions of the paper (regardless of whether the code and da...

  82. [90]

    Guidelines: • The answer NA means that paper does not include experiments requiring code

    Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instruc- tions to faithfully reproduce the main experimental results, as described in supplemental material? 17 Answer: [Yes] Justification: See Github link provided...

  83. [91]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experimental setting/details Question: Does the paper specify all the training and test details (e.g., data splits, hyper- parameters, how they were chosen, type of optimizer, etc.) necessary to understand the results? Answer: [Yes] Justification: See Section 4. Guidelines: • ...

  84. [92]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiment statistical significance Question: Does the paper report error bars suitably and correctly defined or other appropriate information about the statistical significance of the experiments? Answer: [Yes] Justification: See Section 4. Guidelines: • The answer NA means t...

  85. [93]

    Guidelines: • The answer NA means that the paper does not include experiments

    Experiments compute resources Question: For each experiment, does the paper provide sufficient information on the com- puter resources (type of compute workers, memory, time of execution) needed to reproduce the experiments? Answer: [Yes] Justification: See Section 4. Guidelin...

  86. [94]

    Guidelines: • The answer NA means that the authors have not reviewed the NeurIPS Code of Ethics

    Code of ethics Question: Does the research conducted in the paper conform, in every respect, with the NeurIPS Code of Ethics https://neurips.cc/public/EthicsGuidelines? Answer: [Yes] Justification: See Section 5. Guidelines: • The answer NA means that the authors have not revi...

  87. [95]

    Guidelines: • The answer NA means that there is no societal impact of the work performed

    Broader impacts Question: Does the paper discuss both potential positive societal impacts and negative societal impacts of the work performed? Answer: [Yes] Justification: See Section 5. Guidelines: • The answer NA means that there is no societal impact of the work performed. ...

  88. [96]

    Guidelines: • The answer NA means that the paper poses no such risks

    Safeguards Question: Does the paper describe safeguards that have been put in place for responsible release of data or models that have a high risk for misuse (e.g., pretrained language models, image generators, or scraped datasets)? Answer: [Yes] Justification: See Section 5....

  89. [97]

    Guidelines: • The answer NA means that the paper does not use existing assets

    Licenses for existing assets Question: Are the creators or original owners of assets (e.g., code, data, models), used in the paper, properly credited and are the license and terms of use explicitly mentioned and properly respected? Answer: [Yes] Justification: We have provided...

  90. [98]

    Guidelines: • The answer NA means that the paper does not release new assets

    New assets Question: Are new assets introduced in the paper well documented and is the documentation provided alongside the assets? 20 Answer: [Yes] Justification: See Section 4. Guidelines: • The answer NA means that the paper does not release new assets. • Researchers should...

  91. [99]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Crowdsourcing and research with human subjects Question: For crowdsourcing experiments and research with human subjects, does the paper include the full text of instructions given to participants and screenshots, if applicable, as well as details about compensation (if any)? A...

  92. [100]

    Guidelines: • The answer NA means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

  93. [101]

    Answer: [Yes] 21 Justification: We use LLMs for checking the writing errors in the paper

    Declaration of LLM usage Question: Does the paper describe the usage of LLMs if it is an important, original, or non-standard component of the core methods in this research? Note that if the LLM is used only for writing, editing, or formatting purposes and does not impact the ...

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.