Pith. sign in

REVIEW 4 major objections 6 minor 75 references

A Satellite-Ground Synergistic Large Vision-Language Model System for Earth Observation

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read SpaceVerse claims a satellite-ground synergistic LVLM system achieves 31.2% higher accuracy and 51.2% lower latency than state-of-the-art baselines.

desk verdict A plausible first-cut satellite-ground LVLM co-design, but the headline latency number rests on an unstated contact-window assumption in an emulated testbed. read the letter →

arxiv 2507.05731 v1 pith:CU6YHYI7 submitted 2025-07-08 cs.NI cs.AIcs.LG

Yuxin Zhang , Jiahao Yang , Zhe Chen , Wenjun Zhu , Jin Zhao , Yue Gao This is my paper
classification cs.NIcs.AIcs.LG
keywords largevision-languagemodelsLEOsatellitenetworkssatellite-groundsynergisticinferenceEarthobservationtaskoffloadingmulti-scalepreprocessingprogressiveconfidencenetworktext-imageattention
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

This paper tries to establish that a satellite-ground cooperative system can make large vision-language models practical for low-Earth-orbit Earth observation. Its central claim is that by running a compact 2B model on the satellite and a 7B model on the ground, and deciding per image which model should handle it, the system lands on a better accuracy-latency frontier than either endpoint alone. A progressive confidence network makes that decision early, and an attention-guided multi-scale preprocessing step removes redundant image regions before transmission. On three Earth observation benchmarks, the authors report a 31.2% average accuracy gain and a 51.2% latency reduction over state-of-the-art baselines.

What carries the argument

The load-bearing objects are the progressive confidence network and the attention-based multi-scale preprocessing. The confidence network $\tilde{g}$ is an MLP with $I$ linear projections; at iteration $i$ it takes only the visual features $V(x_k)$ and, in later iterations, the onboard model's generated tokens $A_{i-1}$, and compares its output against thresholds $\tau_i$ to decide whether to offload or continue. The preprocessing step computes a text-image attention score $K(x_k^r)$ between each image region's visual features and the text prompt's features, then applies a piecewise filter: discard if $K < \alpha$, downsample on a scale that grows as attention rises when $\alpha \le K < \beta$, and keep full resolution when $K \ge \beta$. The thresholds ($\tau_1=0.5$, $\tau_2=0.4$, $\alpha=0.35$, $\beta=0.55$) are hand-set in the experiments.

What would settle it

Run SpaceVerse on a hardware-in-the-loop testbed with a radiation-hardened flight processor and a real, intermittently visible satellite downlink, or in an actual orbital demonstration, and measure per-sample latency and accuracy across several orbits; if the 31.2% average accuracy gain and 51.2% latency reduction shrink or reverse under real visibility windows, the central claim fails.

Watch

Extended reading notes

Core claim

SpaceVerse's discovery is that the accuracy-latency tradeoff in satellite LVLM inference is best managed by co-designing computation and communication, not by choosing a single deployment point. The progressive confidence network $\tilde{g}$ uses image features alone to flag samples that should go to the ground immediately, then refines the estimate with tokens generated by the onboard model, so hard samples are offloaded without waiting for full onboard inference. Before any offloaded image is transmitted, a text-image attention score $K(x_k^r)$ identifies which regions matter for the prompt, discarding or downsampling the rest. Together these two mechanisms produce the reported gains: 31.2% higher average accuracy and 51.2% lower latency than satellite-only, ground-only, Tabi, and AI-RG baselines across RSVQA, RESISC45, and DOTA-v1.0.

Load-bearing premise

The measured gains rest on the assumption that the emulated satellite platform and downlink faithfully reproduce real low-Earth-orbit conditions; if actual onboard compute or link behavior differs, the reported accuracy and latency numbers may not transfer.

Editorial extensions

If this is right

  • If the reported numbers transfer to operational constellations, near real-time LVLM analysis of Earth observation images, including disaster and extreme weather monitoring, becomes feasible without waiting for full image downloads.
  • Early offloading via the confidence network means hard samples reach the ground model before the onboard model finishes generation, cutting the per-sample compute budget on the satellite.
  • Attention-guided preprocessing lets the system sustain accuracy at high compression ratios: the paper reports only a 4.1% accuracy drop on DOTA at a 5:1 compression ratio, so bandwidth-limited links can carry far less data per image.
  • The single-satellite, single-ground-station formulation is described as readily extensible, so the same co-design could apply to multi-satellite, multi-ground-station settings.
  • The accuracy-latency frontier improves over both endpoints and over existing edge-cloud offloading systems, so system designers no longer have to choose between onboard capability and ground accuracy.

Reading between the lines

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

  • If SpaceVerse's gains hold, the same progressive-confidence-plus-attention-filter pattern could apply to other bandwidth-starved edge-cloud settings, such as UAVs, ships, or ground sensors, wherever a small onboard model and a large remote model share an intermittent link.
  • A testable extension is to make the four hand-set thresholds ($\tau_1$, $\tau_2$, $\alpha$, $\beta$) adaptive per orbit or per task, because the paper fixes them manually and does not show how sensitive the gains are to those values.
  • The single-satellite, single-ground-station evaluation leaves multi-satellite handover and link contention unmeasured; verifying behavior under constellation-level scheduling would be a natural next step.
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 / 6 minor

Summary. SpaceVerse is a satellite-ground synergistic inference system that pairs a compact 2B LVLM deployed on a satellite (emulated on Jetson AGX Xavier) with a 7B LVLM at a ground station. A progressive confidence network uses image features and, progressively, generated tokens to decide whether each sample should be processed onboard or offloaded, and an attention-based multi-scale preprocessor discards or downsamples image regions with low text-image relevance before satellite-to-ground transmission. The authors prototype the satellite-ground link with Open vSwitch/tc shaped by Starlink traces, evaluate on RSVQA, RESISC45, and DOTA-v1.0, and report a 31.2% average accuracy gain and a 51.2% latency reduction over state-of-the-art baselines.

Significance. The paper addresses a genuine and timely deployment gap: bringing LVLMs to LEO Earth-observation systems under tight onboard compute and intermittent connectivity. The progressive confidence network is an interesting idea because it allows early, cheap offload decisions without always waiting for full onboard inference, and the attention-guided multi-scale preprocessing is well motivated by the redundancy analysis in Section 2.2. The authors provide a full prototype implementation and ablation studies that are directionally informative, and the training/evaluation protocol (5% training split for the confidence network, held-out test sets) avoids circularity. If the headline gains survive a more rigorous evaluation, the system would be a useful contribution to satellite-ground collaborative inference. The main weakness is that the quantitative claims currently rest on a single emulated-testbed configuration with no uncertainty characterization and with an unspecified treatment of contact-window visibility.

major comments (4)
  1. [§4.1.1, §4.1.4, Figure 9] The paper never states whether the Open vSwitch/tc emulation restricts transmission to the TLE-derived contact windows, even though Section 2.1.2 reports that contact windows average only 4.33% of the orbital period. If the emulator continuously shapes traffic at 110.67 Mbps rather than gating it by visibility, the latency comparison omits the dominant queuing and visibility delay of real LEO links, and the headline 51.2% latency reduction may not transfer to operational settings. Please state the scheduling assumption explicitly; if contact-window gating is not implemented, rerun the evaluation with it enabled or provide a quantitative bound on its effect.
  2. [Figures 9-12] No error bars, confidence intervals, or multiple-run statistics are reported anywhere in the evaluation. The 31.2% average accuracy gain and the 51.2% latency reduction are single point estimates from one prototype configuration, so it is impossible to assess whether the differences between SpaceVerse and the baselines are significant or stable across runs. Report means and variances over repeated runs (or at least multiple random seeds and data splits) for all headline comparisons.
  3. [§4.1.4, Eq. (3)] The thresholds tau1=0.5, tau2=0.4, alpha=0.35, and beta=0.55 are fixed by hand with no sensitivity analysis. Both the accuracy-latency tradeoff from the confidence network and the compression behavior of the multi-scale preprocessor depend directly on these values, so the claim that the chosen configuration generalizes beyond the three test datasets is unsupported. Please provide a sensitivity study over the threshold ranges or a principled selection procedure, and show how the reported gains vary with the thresholds.
  4. [§3.1.4, Eq. (1)] The training loss defines the target similarity as the cosine similarity between the satellite and ground outputs, but for DOTA-v1.0 the output is a variable-length set of detection boxes and categories. It is not specified how such outputs are converted to a vector for cosine similarity, which is load-bearing for the confidence network's training signal. Please specify the output representation used for object detection, or justify why the cosine formulation is well-defined for all three tasks.
minor comments (6)
  1. [Figure 3 caption] The caption says 'Musk ratios' and the text refers to a 'Musk strategy'; these should be 'Mask ratios' and 'Mask strategy'.
  2. [Abstract, Algorithm 1, Figure 7] There are spacing errors in 'S paceVerse' in the abstract and in several places in the body; these should be fixed for consistency.
  3. [Eq. (1)] The typesetting of Eq. (1) is garbled: the MSE term and the cosine-similarity target are not clearly separated, making the loss definition hard to read.
  4. [Algorithm 1] In Algorithm 1, line 7 assigns y_k from the satellite model outside the if branch; this should be an else branch so that satellite inference is the fallback when no offload decision is made.
  5. [§4.1.2] The paper says '5% train sets are used to update the parameters of g-tilde', but it is unclear whether this means 5% of each dataset's official training split or 5% of the full dataset; please clarify.
  6. [Figure 9 caption] The caption abbreviates RESISC45 as 'RESISC'; please use the full dataset name for consistency with the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the confidence network and attention preprocessing are trained/evaluated on disjoint splits and measured against external baselines; self-citations are background only.

full rationale

SpaceVerse's two components are self-contained against external benchmarks. The progressive confidence network (Section 3.1) is trained by supervised regression (Eq. 1, Section 3.1.4) on 5% train splits to predict the cosine similarity between Qwen2-VL-2B and Qwen2-VL-7B outputs; the reported accuracy (Section 4.2, Figure 9) is measured on held-out test samples, so the offloading 'prediction' is not the same quantity as the evaluated metric and is not forced by construction. The attention-based multi-scale preprocessing (Section 3.2) computes CLIP text-image cosine similarity (Eq. 2) as a relevance signal and applies hand-set thresholds (alpha = 0.35, beta = 0.55) to discard or downsample regions; it does not use the final task accuracy as a training target, nor is the gain an algebraic identity. The latency and accuracy gains are claimed with respect to external baselines (Tabi, AI-RG, satellite-only, GS-only) measured on the same testbed. Self-citations (e.g., refs [4], [5], [6], [35], [54]) appear in the background discussion of LEO networking and split learning but are not invoked as the basis for any uniqueness theorem or as a load-bearing premise for the system's design or evaluation. The skeptical concern about whether the emulator gates transmission by TLE contact windows is a fidelity/validity limitation, not a circularity, and does not affect this assessment.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on several domain assumptions: emulated hardware and links stand in for real satellites, cosine similarity of model outputs is a usable confidence signal, and text-image attention identifies the regions that matter to the ground model. The only tuned quantities are the four thresholds and two structural hyperparameters; no new physical entities are introduced.

free parameters (7)
  • alpha = 0.35
    Attention threshold below which image regions are discarded before downlink; set in §4.1.4 without a sensitivity study.
  • beta = 0.55
    Attention threshold above which image regions are preserved at full resolution; hand-set in §4.1.4.
  • tau_1 = 0.5
    First-stage confidence threshold for offloading; set in §4.1.4.
  • tau_2 = 0.4
    Second-stage confidence threshold for offloading; set in §4.1.4.
  • I = 2
    Number of progressive confidence stages; fixed hyperparameter.
  • N_r^k region granularity = 100
    Number of regions per image for multi-scale preprocessing; set in §4.1.4.
  • N_t tokens per stage
    Number of tokens generated between confidence checks is not reported, but it controls the latency/accuracy trade-off of the progressive network.
assumptions (4)
  • domain assumption Cosine similarity between satellite and ground LVLM outputs is a valid proxy for whether a sample needs ground assistance.
    Used to train the confidence network in §3.1.4; if the 2B model is confidently wrong while still similar to the 7B output in embedding space, the offloading decision would be mis-calibrated.
  • domain assumption Jetson AGX Xavier (or KVM emulation) computational capacity closely resembles real LEO satellite onboard processors.
    Used in §2.1.1 and §4.1.1 to justify model choice and latency measurements; actual flight hardware differs in power, thermal, and radiation constraints.
  • domain assumption Text-image attention scores computed from CLIP or the onboard encoder identify task-relevant image regions.
    Used in §3.2.2 and §3.2.3 to decide which regions to discard or downsample; Figure 3(b) shows random masking is suboptimal, but the paper does not validate attention quality directly.
  • domain assumption Emulated Open vSwitch links shaped by Starlink traces are representative of real satellite-GS connectivity.
    Used in §4.1.1 for all latency measurements; real LEO links have intermittent visibility, Doppler, weather, and contention effects not captured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Satellite-Ground Synergistic Large Vision-Language Model System for Earth Observation." pith.science (2026). https://pith.science/paper/CU6YHYI7

@misc{pith2026250705731,
  author       = {Pith},
  title        = {Pith review of: A Satellite-Ground Synergistic Large Vision-Language Model System for Earth Observation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CU6YHYI7}},
  note         = {Machine review of arXiv:2507.05731}
}
read the original abstract

Recently, large vision-language models (LVLMs) unleash powerful analysis capabilities for low Earth orbit (LEO) satellite Earth observation images in the data center. However, fast satellite motion, brief satellite-ground station (GS) contact windows, and large size of the images pose a data download challenge. To enable near real-time Earth observation applications (e.g., disaster and extreme weather monitoring), we should explore how to deploy LVLM in LEO satellite networks, and design SpaceVerse, an efficient satellite-ground synergistic LVLM inference system. To this end, firstly, we deploy compact LVLMs on satellites for lightweight tasks, whereas regular LVLMs operate on GSs to handle computationally intensive tasks. Then, we propose a computing and communication co-design framework comprised of a progressive confidence network and an attention-based multi-scale preprocessing, used to identify on-satellite inferring data, and reduce data redundancy before satellite-GS transmission, separately. We implement and evaluate SpaceVerse on real-world LEO satellite constellations and datasets, achieving a 31.2% average gain in accuracy and a 51.2% reduction in latency compared to state-of-the-art baselines.

Figures

Figures reproduced from arXiv: 2507.05731 by the authors.

Figure 1
Figure 1. Firstly, satellites’ strict computational constraints in power, size, and weight [51, 52] make deploying reg￾ular LVLMs (e.g., Qwen2-VL-7B [53]) infeasible, allowing 1 arXiv:2507.05731v1 [cs.NI] 8 Jul 2025 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The comparison of performance and memory [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. Intermittent satellite-GS connectivity (a) [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: The SpaceVerse framework overview consists of two key successive components: i) allocating LVLM inference tasks between satellites and GS, and ii) mitigating satellite data redundancy before transmission. onboard, while proposing a progressive confidence network for ef…
Figure 6
Figure 6. Figure 6: The operational mechanism of the onboard [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: SpaceVerse adaptively compresses RS data by leveraging the attention between task prompt and different regions, minimizing satellite-GS transmission overhead while preserving inference performance. processing distinct strategies: preserving the original reso￾lution, do…
Figure 8
Figure 8. Figure 8: SpaceVerse prototype and testbed. 6 [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: The inference latency and performance on RSVQA, RESISC, and DOTA datasets. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 11
Figure 11. Figure 11: Ablation experiments for progressive confi [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 12
Figure 12. Figure 12: Ablation studies and visualization for multi [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

75 extracted references · 59 canonical work pages

  1. [1]

    Democratizing{Direct- to-Cell} Low Earth Orbit Satellite Networks

    Lixin Liu, Yuanjie Li, Hewu Li, Jiabo Yang, Wei Liu, Jingyi Lan, Yufeng Wang, Jiarui Li, Jianping Wu, Qian Wu, et al. Democratizing{Direct- to-Cell} Low Earth Orbit Satellite Networks. In Proc. of the 21st NSDI , pages 791–808, 2024

    work page 2024

  2. [2]

    Robust Live Stream- ing over LEO Satellite Constellations: Measurement, Analysis, and Handover-Aware Adaptation

    Hao Fang, Haoyuan Zhao, Jianxin Shi, Miao Zhang, Guanzhen Wu, Yi Ching Chou, Feng Wang, and Jiangchuan Liu. Robust Live Stream- ing over LEO Satellite Constellations: Measurement, Analysis, and Handover-Aware Adaptation. In Proc. of the 32nd ACM MM , pages 5958–5966, 2024

    work page 2024

  3. [3]

    Spectrumize: Spectrum-Efficient Satellite Networks for the Internet of Things

    Vaibhav Singh, Tusher Chakraborty, Suraj Jog, Om Chabra, Deepak Va- sisht, and Ranveer Chandra. Spectrumize: Spectrum-Efficient Satellite Networks for the Internet of Things. In Proc. of the 21st NSDI , pages 825–840, 2024

    work page 2024

  4. [4]

    SatFed: A Resource-Efficient LEO Satellite-Assisted Heterogeneous Federated Learning Framework

    Yuxin Zhang, Zheng Lin, Zhe Chen, Zihan Fang, Wenjun Zhu, Xi- anhao Chen, Jin Zhao, and Yue Gao. SatFed: A Resource-Efficient LEO Satellite-Assisted Heterogeneous Federated Learning Framework. arXiv preprint arXiv:2409.13503, 2024

    arXiv 2024

  5. [5]

    LEO-Split: A Semi-Supervised Split Learning Framework over LEO Satellite Networks

    Zheng Lin, Yuxin Zhang, Zhe Chen, Zihan Fang, Cong Wu, Xianhao Chen, Yue Gao, and Jun Luo. LEO-Split: A Semi-Supervised Split Learning Framework over LEO Satellite Networks. arXiv preprint arXiv:2501.01293, 2025

    arXiv 2025

  6. [6]

    ESL-LEO: An Efficient Split Learning Framework over LEO Satellite Networks

    Zheng Lin, Yuxin Zhang, Zhe Chen, Zihan Fang, Yanni Yang, Guoming Zhang, Huan Yang, Cong Wu, Xianhao Chen, and Yue Gao. ESL-LEO: An Efficient Split Learning Framework over LEO Satellite Networks. In Proc. Int. Conf. Wireless Artif. Intell. Comput. Syst. Appl. , pages 344–357, 2025

    work page 2025

  7. [7]

    The Digital Divide in Canada and the Role of LEO Satellites in Bridging the Gap.IEEE Commun

    Tuheen Ahmmed, Afsoon Alidadi, Zichao Zhang, Aizaz U Chaudhry, and Halim Yanikomeroglu. The Digital Divide in Canada and the Role of LEO Satellites in Bridging the Gap.IEEE Commun. Mag., 60(6):24–30, 2022

    work page 2022

  8. [8]

    Available: https://www.planet.com/

    “Planet", 2024. Available: https://www.planet.com/

    work page 2024

Show all 75 references
  1. [9]

    SatSense: Multi-Satellite Collabo- rative Framework for Spectrum Sensing

    Haoxuan Yuan, Zhe Chen, Zheng Lin, Jinbo Peng, Zihan Fang, Yuhang Zhong, Zihang Song, and Yue Gao. SatSense: Multi-Satellite Collabo- rative Framework for Spectrum Sensing. IEEE Trans. Cogn. Commun. Netw., 2025

    2025

  2. [10]

    SigChord: Sniffing Wide Non-Sparse Multiband Signals for Terrestrial and Non-Terrestrial Wireless Networks

    Jinbo Peng, Junwen Duan, Zheng Lin, Haoxuan Yuan, Yue Gao, and Zhe Chen. SigChord: Sniffing Wide Non-Sparse Multiband Signals for Terrestrial and Non-Terrestrial Wireless Networks. arXiv preprint arXiv:2504.06587, 2025

    2025 arXiv

  3. [11]

    FedSN: A Federated Learning Framework over Heteroge- neous LEO Satellite Networks

    Zheng Lin, Zhe Chen, Zihan Fang, Xianhao Chen, Xiong Wang, and Yue Gao. FedSN: A Federated Learning Framework over Heteroge- neous LEO Satellite Networks. IEEE Trans. Mobile Comput. , 2024

    2024

  4. [12]

    LEO Satellite Networks Assisted Geo-Distributed Data Processing

    Zhiyuan Zhao, Zhe Chen, Zheng Lin, Wenjun Zhu, Kun Qiu, Chaoqun You, and Yue Gao. LEO Satellite Networks Assisted Geo-Distributed Data Processing. IEEE Wireless Commun. Lett. , 2024

    2024

  5. [13]

    A Networking Perspective on Starlink’s Self-Driving LEO Mega-Constellation

    Yuanjie Li, Hewu Li, Wei Liu, Lixin Liu, Wei Zhao, Yimei Chen, Jianping Wu, Qian Wu, Jun Liu, Zeqi Lai, et al. A Networking Perspective on Starlink’s Self-Driving LEO Mega-Constellation. In Proc. of the 29th ACM MobiCom, pages 1–16, 2023

    2023

  6. [14]

    S4: Self-Supervised Sensing Across the Spectrum

    Jayanth Shenoy, Xinjian Davis Zhang, Shlok Mehrotra, Bill Tao, Rem Yang, Han Zhao, and Deepak Vasisht. S4: Self-Supervised Sensing Across the Spectrum. arXiv preprint arXiv:2405.01656, 2024

    2024 arXiv

  7. [15]

    Utilizing Very High-resolution Optical RGB Satellite Imagery in Geo-information Extraction for Fine-scale Map-making

    Wenmiao Hu. Utilizing Very High-resolution Optical RGB Satellite Imagery in Geo-information Extraction for Fine-scale Map-making. In Proc. of the 32nd ACM MM , pages 11127–11131, 2024

    2024

  8. [16]

    Artificial Intelligence Revolutionises Weather Forecast, Climate Moni- toring and Decadal Prediction

    Steven Dewitte, Jan P Cornelis, Richard Müller, and Adrian Munteanu. Artificial Intelligence Revolutionises Weather Forecast, Climate Moni- toring and Decadal Prediction. Remote Sensing, 13(16):3209, 2021

    2021

  9. [17]

    Impacts of Climate Variability and Drought on Surface Water Resources in Sub-Saharan Africa Using Remote Sensing: A Review

    Trisha Deevia Bhaga, Timothy Dube, Munyaradzi Davis Shekede, and Cletah Shoko. Impacts of Climate Variability and Drought on Surface Water Resources in Sub-Saharan Africa Using Remote Sensing: A Review. Remote Sensing, 12(24):4184, 2020

    2020

  10. [18]

    Basic Performance and Future Developments of BeiDou Global Navigation Satellite System

    Yuanxi Yang, Yue Mao, and Bijiao Sun. Basic Performance and Future Developments of BeiDou Global Navigation Satellite System. Satellite Navigation, 1(1):1, 2020

    2020

  11. [19]

    Simultaneous Localization and Mapping (SLAM) for Au- tonomous Driving: Concept and Analysis

    Shuran Zheng, Jinling Wang, Chris Rizos, Weidong Ding, and Ahmed El-Mowafy. Simultaneous Localization and Mapping (SLAM) for Au- tonomous Driving: Concept and Analysis. Remote Sensing, 15(4):1156, 2023

    2023

  12. [20]

    Spatial Analysis and GIS in the Study of COVID-19

    Ivan Franch-Pardo, Brian M Napoletano, Fernando Rosete-Verges, and Lawal Billa. Spatial Analysis and GIS in the Study of COVID-19. A Review. Sci. Total Environ., 739:140033, 2020

    2020

  13. [21]

    Seeing Through Clouds in Satellite Images

    Mingmin Zhao, Peder Olsen, and Ranveer Chandra. Seeing Through Clouds in Satellite Images. IEEE Trans. Geosci. Remote Sens. , 61:1–16, 2023

    2023

  14. [22]

    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. Nat. Commun., 12(1):4392, 2021

    2021

  15. [23]

    Large Selective Kernel Network for Remote Sensing Object Detection

    Yuxuan Li, Qibin Hou, Zhaohui Zheng, Ming-Ming Cheng, Jian Yang, and Xiang Li. Large Selective Kernel Network for Remote Sensing Object Detection. In Proc. of the 18th IEEE/CVF ICCV , pages 16794– 16805, 2023

    2023

  16. [24]

    Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks

    Zheng Lin, Guangyu Zhu, Yiqin Deng, Xianhao Chen, Yue Gao, Kaibin Huang, and Yuguang Fang. Efficient Parallel Split Learning over Resource-Constrained Wireless Edge Networks. IEEE Trans. Mobile Comput., 23(10):9224–9239, 2024

    2024

  17. [25]

    IC3M: In-Car Multimodal Multi- Object Monitoring for Abnormal Status of Both Driver and Passengers

    Zihan Fang, Zheng Lin, Senkang Hu, Hangcheng Cao, Yiqin Deng, Xianhao Chen, and Yuguang Fang. IC3M: In-Car Multimodal Multi- Object Monitoring for Abnormal Status of Both Driver and Passengers. arXiv preprint arXiv:2410.02592, 2024

    arXiv

  18. [26]

    RF-Based Human Activity Recognition Using Signal Adapted 9 ACM MM’25, October, 2025, Dublin, Ireland Y

    Zhe Chen, Chao Cai, Tianyue Zheng, Jun Luo, Jie Xiong, and Xin Wang. RF-Based Human Activity Recognition Using Signal Adapted 9 ACM MM’25, October, 2025, Dublin, Ireland Y. Zhang, J. Yang, Z. Chen, W, Zhu, J Zhao, and Y. Gao Convolutional Neural Network. IEEE Trans. Mobile Com...

    2025

  19. [27]

    SUMS: Sniffing Unknown Multiband Signals under Low Sampling Rates

    Jinbo Peng, Zhe Chen, Zheng Lin, Haoxuan Yuan, Zihan Fang, Lingzhong Bao, Zihang Song, Ying Li, Jing Ren, and Yue Gao. SUMS: Sniffing Unknown Multiband Signals under Low Sampling Rates. IEEE Trans. Mobile Comput., 2024

    2024

  20. [28]

    Constructing 4D Radio Map in LEO Satellite Networks with Limited Samples

    Haoxuan Yuan, Zhe Chen, Zheng Lin, Jinbo Peng, Yuhang Zhong, Xuanjie Hu, Songyan Xue, Wei Li, and Yue Gao. Constructing 4D Radio Map in LEO Satellite Networks with Limited Samples. IEEE INFOCOM, 2025

    2025

  21. [29]

    Accelerating Federated Learning with Model Segmentation for Edge Networks

    Mingda Hu, Jingjing Zhang, Xiong Wang, Shengyun Liu, and Zheng Lin. Accelerating Federated Learning with Model Segmentation for Edge Networks. IEEE Trans. Green Commun. Netw., 2024

    2024

  22. [30]

    MERIT: Multimodal Wearable Vital Sign Waveform Monitoring

    Yongyang Tang, Zhe Chen, Ang Li, Tianyue Zheng, Zheng Lin, Jia Xu, Pin Lv, Zhe Sun, and Yue Gao. MERIT: Multimodal Wearable Vital Sign Waveform Monitoring. arXiv preprint arXiv:2410.00392, 2024

    2024 arXiv

  23. [31]

    Convolutional Neural Networks Based Remote Sensing Scene Classification under Clear and Cloudy Environments

    Huiming Sun, Yuewei Lin, Qin Zou, Shaoyue Song, Jianwu Fang, and Hongkai Yu. Convolutional Neural Networks Based Remote Sensing Scene Classification under Clear and Cloudy Environments. In Proc. of the 34th IEEE/CVF CVPR , pages 713–720, 2021

    2021

  24. [32]

    Graph Learning for Multi-Satellite Based Spectrum Sensing

    Haoxuan Yuan, Zhe Chen, Zheng Lin, Jinbo Peng, Zihan Fang, Yuhang Zhong, Zihang Song, Xiong Wang, and Yue Gao. Graph Learning for Multi-Satellite Based Spectrum Sensing. In Proc. IEEE Int. Conf. Commun. Technol. (ICCT), pages 1112–1116, 2023

    2023

  25. [33]

    HASFL: Heterogeneity-Aware Split Federated Learning over Edge Computing Systems

    Zheng Lin, Zhe Chen, Xianhao Chen, Wei Ni, and Yue Gao. HASFL: Heterogeneity-Aware Split Federated Learning over Edge Computing Systems. arXiv preprint arXiv:2506.08426, 2025

    2025 arXiv

  26. [34]

    Netllm: Adapting Large Language Models for Networking

    Duo Wu, Xianda Wang, Yaqi Qiao, Zhi Wang, Junchen Jiang, Shuguang Cui, and Fangxin Wang. Netllm: Adapting Large Language Models for Networking. In Proc. of the 34th ACM SIGCOMM , pages 661–678, 2024

    2024

  27. [35]

    LCFed: An Efficient Clustered Federated Learning Framework for Heteroge- neous Data

    Yuxin Zhang, Haoyu Chen, Zheng Lin, Zhe Chen, and Jin Zhao. LCFed: An Efficient Clustered Federated Learning Framework for Heteroge- neous Data. arXiv preprint arXiv:2501.01850, 2025

    2025 arXiv

  28. [36]

    Gradient free personalized federated learning

    Haoyu Chen, Yuxin Zhang, Jin Zhao, Xin Wang, and Yuedong Xu. Gradient free personalized federated learning. In Proceedings of the 53rd International Conference on Parallel Processing , pages 971–980, 2024

    2024

  29. [37]

    Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks

    Zheng Lin, Guanqiao Qu, Wei Wei, Xianhao Chen, and Kin K Leung. Adaptsfl: Adaptive Split Federated Learning in Resource-Constrained Edge Networks. IEEE Trans. Netw., 2024

    2024

  30. [38]

    Scaling Laws for Neural Language Models

    Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Ben- jamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling Laws for Neural Language Models. arXiv preprint arXiv:2001.08361, 2020

    2001 arXiv

  31. [39]

    SpectralGPT: Spectral Remote Sensing Foundation Model

    Danfeng Hong, Bing Zhang, Xuyang Li, Yuxuan Li, Chenyu Li, Jing Yao, Naoto Yokoya, Hao Li, Pedram Ghamisi, Xiuping Jia, Antonio Plaza, Paolo Gamba, Jon Atli Benediktsson, and Jocelyn Chanussot. SpectralGPT: Spectral Remote Sensing Foundation Model. IEEE Trans. Pattern Anal. Ma...

    2024

  32. [40]

    HSplitLoRA: A Heterogeneous Split Parameter-Efficient Fine-Tuning Framework for Large Language Models

    Zheng Lin, Yuxin Zhang, Zhe Chen, Zihan Fang, Xianhao Chen, Pra- neeth Vepakomma, Wei Ni, Jun Luo, and Yue Gao. HSplitLoRA: A Heterogeneous Split Parameter-Efficient Fine-Tuning Framework for Large Language Models. arXiv preprint arXiv:2505.02795, 2025

    2025 arXiv

  33. [41]

    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 Trans. Geosci. Remote Sens., 2024

    2024

  34. [42]

    EarthGPT: A Universal Multimodal Large Language Model for Multi- sensor Image Comprehension in Remote Sensing Domain

    Wei Zhang, Miaoxin Cai, Tong Zhang, Yin Zhuang, and Xuerui Mao. EarthGPT: A Universal Multimodal Large Language Model for Multi- sensor Image Comprehension in Remote Sensing Domain. IEEE Trans. Geosci. Remote Sens., 62:1–20, 2024

    2024

  35. [43]

    SkyEyeGPT: Unifying Remote Sensing Vision-Language Tasks via Instruction Tuning with Large Language Model

    Yang Zhan, Zhitong Xiong, and Yuan Yuan. SkyEyeGPT: Unifying Remote Sensing Vision-Language Tasks via Instruction Tuning with Large Language Model. arXiv preprint arXiv:2401.09712, 2024

    2024 arXiv

  36. [44]

    Vision- Language Models for Vision Tasks: A Survey

    Jingyi Zhang, Jiaxing Huang, Sheng Jin, and Shijian Lu. Vision- Language Models for Vision Tasks: A Survey. IEEE Trans. Pattern Anal. Mach. Intell., 46(8):5625–5644, 2024

    2024

  37. [45]

    PIP: Detecting Adversarial Examples in Large Vision- Language Models via Attention Patterns of Irrelevant Probe Questions

    Yudong Zhang, Ruobing Xie, Jiansheng Chen, Xingwu Sun, and Yu Wang. PIP: Detecting Adversarial Examples in Large Vision- Language Models via Attention Patterns of Irrelevant Probe Questions. In Proc. of the 32nd ACM MM , pages 11175–11183, 2024

    2024

  38. [46]

    Break the Visual Perception: Adversarial Attacks Targeting Encoded Visual Tokens of Large Vision-Language Models

    Yubo Wang, Chaohu Liu, Yanqiu Qu, Haoyu Cao, Deqiang Jiang, and Linli Xu. Break the Visual Perception: Adversarial Attacks Targeting Encoded Visual Tokens of Large Vision-Language Models. In Proc. of the 32nd ACM MM, pages 1072–1081, 2024

    2024

  39. [47]

    Automated Federated Pipeline for Parameter- Efficient Fine-Tuning of Large Language Models

    Zihan Fang, Zheng Lin, Zhe Chen, Xianhao Chen, Yue Gao, and Yuguang Fang. Automated Federated Pipeline for Parameter- Efficient Fine-Tuning of Large Language Models. arXiv preprint arXiv:2404.06448, 2024

    2024 arXiv

  40. [48]

    SplitLoRA: A Split Parameter-Efficient Fine-Tuning Framework for Large Language Models

    Zheng Lin, Xuanjie Hu, Yuxin Zhang, Zhe Chen, Zihan Fang, Xianhao Chen, Ang Li, Praneeth Vepakomma, and Yue Gao. SplitLoRA: A Split Parameter-Efficient Fine-Tuning Framework for Large Language Models. arXiv preprint arXiv:2407.00952, 2024

    2024 arXiv

  41. [49]

    Zero-Shot Text-to-Image Generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-Shot Text-to-Image Generation. In Proc. of the 38th ICML , pages 8821–8831, 2021

    2021

  42. [50]

    GeoChat: Grounded Large Vision-Language Model for Remote Sensing

    Kartik Kuckreja, Muhammad Sohail Danish, Muzammal Naseer, Abhijit Das, Salman Khan, and Fahad Shahbaz Khan. GeoChat: Grounded Large Vision-Language Model for Remote Sensing. In Proc. of the 37th IEEE/CVF CVPR, pages 27831–27840, 2024

    2024

  43. [51]

    Orbital Edge Computing: Nanosatellite Constellations as a New Class of Computer System

    Bradley Denby and Brandon Lucia. Orbital Edge Computing: Nanosatellite Constellations as a New Class of Computer System. In Proc. of the 25th ACM ASPLOS , pages 939–954, 2020

    2020

  44. [52]

    Small Satellites and Big Antennas

    “Small Satellites and Big Antennas", 2023. Available: https://www.raspberrypi.com/news/small-satellites-and-big- antennas

    2023

  45. [53]

    Qwen2-VL: Enhancing Vision-Language Model’s Perception of the World at Any Resolution

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, et al. Qwen2-VL: Enhancing Vision-Language Model’s Perception of the World at Any Resolution. arXiv preprint arXiv:2409.12191, 2024

    2024 arXiv

  46. [54]

    S-leon: An efficient split learning framework over heterogeneous leo satellite networks

    Yuxin Zhang, Zhe Chen, Xuanjie Hu, Jin Zhao, and Yue Gao. S-leon: An efficient split learning framework over heterogeneous leo satellite networks. Authorea Preprints, 2025

    2025

  47. [55]

    L2D2: Low Latency Distributed Downlink for LEO Satellites

    Deepak Vasisht, Jayanth Shenoy, and Ranveer Chandra. L2D2: Low Latency Distributed Downlink for LEO Satellites. In Proc. of the 35th ACM SIGCOMM, pages 151–164, 2021

    2021

  48. [56]

    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 Proc. of the 31st IEEE CVPR , pages 3974–3983, 2018

    2018

  49. [57]

    Deploying Machine Learning Anom- aly Detection Models to Flight Ready AI Boards

    James Murphy, Maria Buckley, Leonie Buckley, Adam Taylor, Jake O’brien, and Brian Mac Namee. Deploying Machine Learning Anom- aly Detection Models to Flight Ready AI Boards. In Proc. of the 41st IEEE/CVF CVPR, pages 6828–6836, 2024

    2024

  50. [58]

    Machine-Learning Space Applications on SmallSat Platforms with TensorFlow

    Jacob Manning, David Langerman, Barath Ramesh, Evan Gretok, Christopher Wilson, Alan George, James MacKinnon, and Gary Crum. Machine-Learning Space Applications on SmallSat Platforms with TensorFlow. 2018

    2018

  51. [59]

    Onboard Processing With Hybrid and Reconfigurable Computing on Small Satellites

    Alan D George and Christopher M Wilson. Onboard Processing With Hybrid and Reconfigurable Computing on Small Satellites. Proc. IEEE, 106(3):458–470, 2018

    2018

  52. [60]

    Peng Xu, Wenqi Shao, Kaipeng Zhang, Peng Gao, Shuo Liu, Meng Lei, Fanqing Meng, Siyuan Huang, Yu Qiao, and Ping Luo. LVLM-EHub: A Comprehensive Evaluation Benchmark for Large Vision-Language 10 A Satellite-Ground Synergistic Large Vision-Language Model System for Earth Observa...

    2025

  53. [61]

    RSVQA: Visual Question Answering for Remote Sensing Data

    Sylvain Lobry, Diego Marcos, Jesse Murray, and Devis Tuia. RSVQA: Visual Question Answering for Remote Sensing Data. IEEE Trans. Geosci. Remote Sens., 58(12):8555–8566, 2020

    2020

  54. [62]

    Remote Sensing Image Scene Classification: Benchmark and State of the Art

    Gong Cheng, Junwei Han, and Xiaoqiang Lu. Remote Sensing Image Scene Classification: Benchmark and State of the Art. Proc. IEEE , 105(10):1865–1883, 2017

    2017

  55. [63]

    Planet Labs PBC Announces Real-Time Insights Tech- nology Using NVIDIA Jetson Platform

    “Planet Labs PBC Announces Real-Time Insights Tech- nology Using NVIDIA Jetson Platform", 2024. Available: https://www.businesswire.com/news/home/20240610385569/en/Planet- Labs-PBC-Announces-Real-Time-Insights-Technology-Using- NVIDIA-Jetson-Platform

    2024

  56. [64]

    Trans- mitting, Fast and Slow: Scheduling Satellite Traffic Through Space and Time

    Bill Tao, Maleeha Masood, Indranil Gupta, and Deepak Vasisht. Trans- mitting, Fast and Slow: Scheduling Satellite Traffic Through Space and Time. In Proc. of the 29th ACM MobiCom , pages 1–15, 2023

    2023

  57. [65]

    NORAD GP Element Sets

    “NORAD GP Element Sets", 2024. Available: https://celestrak.org/NORAD/elements/

    2024

  58. [66]

    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 Proc. of the 32nd ACM MM , pages 6307–6315, 2024

    2024

  59. [67]

    The Design and Implementation of Open vSwitch

    Ben Pfaff, Justin Pettit, Teemu Koponen, Ethan Jackson, Andy Zhou, Jarno Rajahalme, Jesse Gross, Alex Wang, Joe Stringer, Pravin Shelar, et al. The Design and Implementation of Open vSwitch. In Proc. of th 12th NSDI, pages 117–130, 2015

    2015

  60. [68]

    On the Fidelity of Single-Machine Network Emulation in Linux

    Joseph D Beshay, Andrea Francini, and Ravi Prakash. On the Fidelity of Single-Machine Network Emulation in Linux. In Proc. of the 23rd IEEE MASCOTS, pages 19–22, 2015

    2015

  61. [69]

    Learning Transferable Visual Models From Natural Language Supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning Transferable Visual Models From Natural Language Supervision. In Proc. of the 38th ICML , pages 8748–8763, 2021

    2021

  62. [70]

    Tabi: An Efficient Multi-Level Inference System for Large Language Models

    Yiding Wang, Kai Chen, Haisheng Tan, and Kun Guo. Tabi: An Efficient Multi-Level Inference System for Large Language Models. In Proc. of the 18th ECCS EuroSys , pages 233–248, 2023

    2023

  63. [71]

    Large Language Models (LLMs) Inference Offloading and Resource Allocation in Cloud-Edge Computing: An Active Inference Approach

    Ying He, Jingcheng Fang, F Richard Yu, and Victor C Leung. Large Language Models (LLMs) Inference Offloading and Resource Allocation in Cloud-Edge Computing: An Active Inference Approach. IEEE Trans. Mob. Comput., 23(12):11253–11264, 2024

    2024

  64. [72]

    RSGPT: A Remote Sensing Vision Language Model and Benchmark

    Yuan Hu, Jianlong Yuan, Congcong Wen, Xiaonan Lu, and Xiang Li. RSGPT: A Remote Sensing Vision Language Model and Benchmark. arXiv preprint arXiv:2307.15266, 2023

    2023 arXiv

  65. [73]

    Edge- Cloud Polarization and Collaboration: A Comprehensive Survey for AI

    Jiangchao Yao, Shengyu Zhang, Yang Yao, Feng Wang, Jianxin Ma, Jianwei Zhang, Yunfei Chu, Luo Ji, Kunyang Jia, Tao Shen, et al. Edge- Cloud Polarization and Collaboration: A Comprehensive Survey for AI. IEEE Trans. Knowl. Data Eng. , 35(7):6866–6886, 2022

    2022

  66. [74]

    FedAC: An Adaptive Clustered Federated Learning Framework for Heteroge- neous Data

    Yuxin Zhang, Haoyu Chen, Zheng Lin, Zhe Chen, and Jin Zhao. FedAC: An Adaptive Clustered Federated Learning Framework for Heteroge- neous Data. arXiv preprint arXiv:2403.16460, 2024

    2024 arXiv

  67. [75]

    Petals: Collaborative Inference and Fine-tuning of Large Models

    Alexander Borzunov, Dmitry Baranchuk, Tim Dettmers, Max Ryabinin, Younes Belkada, Artem Chumachenko, Pavel Samygin, and Colin Raf- fel. Petals: Collaborative Inference and Fine-tuning of Large Models. arXiv preprint arXiv:2209.01188, 2022. 11

    2022 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.