pith. machine review for the scientific record. sign in

arxiv: 2601.19640 · v2 · submitted 2026-01-27 · 💻 cs.CV

Recognition: unknown

Focus on What Really Matters in Low-Altitude Governance: A Management-Centric Multi-Modal Benchmark with Implicitly Coordinated Vision-Language Reasoning Framework

Hao Chang , Zhihui Wang , Lingxiang Wu , Wei An , Boyang Li , Zaiping Lin , Weidong Sheng , Jinqiao Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-16 11:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords low-altitude visionmulti-modal benchmarkvision-language reasoningaerial perceptiongrounding adaptermanagement-centricsmart city governanceanomaly understanding
0
0 comments X

The pith

GovLA-Reasoner uses a Spatially-aware Grounding Adapter to coordinate fine-grained visual details with language reasoning for low-altitude governance without task-specific fine-tuning.

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

The paper establishes a management-centric benchmark GovLA-10K that annotates only functionally salient targets in aerial images rather than exhaustively labeling all visible objects, and supplies corresponding actionable management suggestions. It introduces GovLA-Reasoner, a unified framework whose Spatially-aware Grounding Adapter compresses multi-stream visual representations and feeds them implicitly into a large language model for integrated reasoning. This design targets practical urban governance needs such as anomaly understanding in smart cities. A sympathetic reader would care because existing object-centric and loosely coupled vision-language pipelines struggle to deliver decision-relevant outputs from low-altitude data.

Core claim

GovLA-10K deliberately centers annotation on targets that map directly to management needs instead of all visible objects and supplies grounded suggestions; GovLA-Reasoner employs the Spatially-aware Grounding Adapter to compress and aggregate grounding-aware representations so that fine-grained spatial cues are preserved and integrated into language reasoning, yielding performance gains while avoiding fine-tuning of any task-specific components.

What carries the argument

The Spatially-aware Grounding Adapter (SGA), which aggregates multi-stream grounding representations from the visual detector and routes compressed spatial cues into the language model for implicit coordination.

Load-bearing premise

The Spatially-aware Grounding Adapter can implicitly coordinate fine-grained visual grounding with high-level language reasoning without task-specific fine-tuning or explicit alignment losses.

What would settle it

If a standard vision-language baseline without the SGA matches or exceeds GovLA-Reasoner accuracy on the management-oriented tasks in GovLA-10K, the claim that the adapter is required for the reported gains would be falsified.

Figures

Figures reproduced from arXiv: 2601.19640 by Boyang Li, Hao Chang, Jinqiao Wang, Lingxiang Wu, Wei An, Weidong Sheng, Zaiping Lin, Zhihui Wang.

Figure 2
Figure 2. Figure 2: Basic statistics of GovLA-10K. (a) Category-wise instance counts and proportions. (b) High-frequency word distribution in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of image numbers across different caption [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Low-altitude framework overview of the existing pipeline (left) and our proposed GovLA-Reasoner (right). To address the infor Stage Ⅱ [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
read the original abstract

Low-altitude vision systems are becoming a critical infrastructure for smart city governance. However, existing object-centric perception paradigms and loosely coupled vision-language pipelines are still difficult to support management-oriented anomaly understanding required in real-world urban governance. To bridge this gap, we introduce GovLA-10K, the first management-oriented multi-modal benchmark for low-altitude intelligence, along with GovLA-Reasoner, a unified vision-language reasoning framework tailored for governance-aware aerial perception. Unlike existing studies that aim to exhaustively annotate all visible objects, GovLA-10K is deliberately designed around functionally salient targets that directly correspond to practical management needs, and further provides actionable management suggestions grounded in these observations. To effectively coordinate the fine-grained visual grounding with high-level contextual language reasoning, GovLA-Reasoner introduces an efficient Spatially-aware Grounding Adapter (SGA) that implicitly coordinates discriminative representation sharing between the visual detector and the large language model (LLM). Different from existing adapters that primarily focus on global embedding alignment, our SGA is specifically designed to compress and aggregate multi-stream grounding-aware representations, thereby preserving fine-grained spatial cues while enabling their effective integration into the language reasoning process. Extensive experiments indicate that our GovLA-Reasoner effectively improves performance while avoiding the need of fine-tuning for any task-specific individual components. We believe our work offers a new perspective and foundation for future studies on management-aware low-altitude vision-language systems. The code and dataset will be publicly released after further organization.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper introduces GovLA-10K, the first management-oriented multi-modal benchmark for low-altitude aerial perception that prioritizes functionally salient targets tied to urban governance needs rather than exhaustive object annotation, and proposes GovLA-Reasoner, a vision-language framework that uses a Spatially-aware Grounding Adapter (SGA) to compress multi-stream representations and implicitly coordinate fine-grained visual grounding with high-level LLM reasoning without task-specific fine-tuning of the detector or LLM.

Significance. If the central claims on performance gains and zero fine-tuning of base components hold under rigorous evaluation, the work would provide a valuable new benchmark and adapter design for management-aware low-altitude systems, shifting focus from generic perception to actionable governance outputs and potentially influencing smart-city applications.

major comments (2)
  1. [Abstract] Abstract: the claim that 'GovLA-Reasoner effectively improves performance while avoiding the need of fine-tuning for any task-specific individual components' is unsupported by any quantitative results, baselines, ablation studies, or error analysis, so the central empirical contribution cannot be assessed for soundness or attribution to the SGA mechanism.
  2. [Methods (SGA)] SGA description and training protocol: the assertion that the Spatially-aware Grounding Adapter implicitly coordinates fine-grained spatial cues with the frozen LLM via compression of multi-stream representations, without explicit alignment losses or updates to the detector/LLM, is load-bearing for the 'no fine-tuning' claim but lacks specification of loss terms, which modules are updated, and the exact training procedure, preventing verification that gains are due to the implicit mechanism rather than hidden supervision.
minor comments (2)
  1. [Methods] Clarify notation for multi-stream representations and compression operations in the SGA to ensure reproducibility.
  2. [Introduction] Add explicit references to related work on vision-language adapters and low-altitude benchmarks in the introduction for better context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment point by point below and have revised the manuscript to strengthen the presentation of our empirical results and methodological details.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'GovLA-Reasoner effectively improves performance while avoiding the need of fine-tuning for any task-specific individual components' is unsupported by any quantitative results, baselines, ablation studies, or error analysis, so the central empirical contribution cannot be assessed for soundness or attribution to the SGA mechanism.

    Authors: We acknowledge that the abstract, as currently written, summarizes the performance claim without embedding specific metrics. The full manuscript reports quantitative results, baseline comparisons, SGA ablations, and error analysis in Sections 4 and 5 that support the claim. To make the abstract self-contained and allow immediate assessment of the central contribution, we will revise it to include key quantitative gains (e.g., accuracy and efficiency improvements) and a concise reference to the experimental validation of the no-fine-tuning property. revision: yes

  2. Referee: [Methods (SGA)] SGA description and training protocol: the assertion that the Spatially-aware Grounding Adapter implicitly coordinates fine-grained spatial cues with the frozen LLM via compression of multi-stream representations, without explicit alignment losses or updates to the detector/LLM, is load-bearing for the 'no fine-tuning' claim but lacks specification of loss terms, which modules are updated, and the exact training procedure, preventing verification that gains are due to the implicit mechanism rather than hidden supervision.

    Authors: We agree that the current Methods section requires greater specificity to allow verification of the implicit coordination mechanism. In the revised manuscript we will expand the SGA subsection to (i) enumerate all loss terms employed during adapter training, (ii) explicitly state that only SGA parameters are updated while the detector and LLM remain frozen, and (iii) provide the complete training protocol, including optimizer settings, batch sizes, and data-flow details. These additions will clarify that no task-specific fine-tuning or hidden supervision is applied to the base components. revision: yes

Circularity Check

0 steps flagged

No circularity: new benchmark and adapter architecture presented as independent contributions

full rationale

The paper introduces GovLA-10K benchmark and GovLA-Reasoner framework with Spatially-aware Grounding Adapter as novel elements. No equations, derivations, or fitted parameters are shown that reduce predictions to inputs by construction. Performance claims rest on experiments rather than self-referential definitions or self-citation chains. The core claims about implicit coordination without task-specific fine-tuning are architectural and empirical, not tautological. This is a standard non-circular presentation of a new dataset and method.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No explicit free parameters, axioms, or invented entities are described in the abstract; the framework relies on standard vision-language components plus the new adapter.

pith-pipeline@v0.9.0 · 5603 in / 1067 out tokens · 24912 ms · 2026-05-16T11:00:23.520460+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 9 internal anchors

  1. [1]

    Multi class aerial im- age classification in uav networks employing snake opti- mization algorithm with deep learning.Scientific Reports, 15(1):23872,

    [Al Mazroaet al., 2025 ] Alanoud Al Mazroa, Nuha Al- ruwais, Muhammad Kashif Saeed, Kamal M Othman, Randa Allafi, and Ahmed S Salama. Multi class aerial im- age classification in uav networks employing snake opti- mization algorithm with deep learning.Scientific Reports, 15(1):23872,

    work page 2025

  2. [2]

    Integrated neural network framework for multi-object detection and recog- nition using uav imagery.Frontiers in neurorobotics, 19:1643011,

    [Alshehriet al., 2025 ] Mohammed Alshehri, Tingting Xue, Ghulam Mujtaba, Yahya AlQahtani, Nouf Abdullah Al- mujally, Ahmad Jalal, and Hui Liu. Integrated neural network framework for multi-object detection and recog- nition using uav imagery.Frontiers in neurorobotics, 19:1643011,

    work page 2025

  3. [3]

    Llava-onevision-1.5: Fully open framework for democratized multimodal training

    [Anet al., 2025 ] Xiang An, Yin Xie, Kaicheng Yang, Wenkang Zhang, Xiuwei Zhao, Zheng Cheng, Yirui Wang, Songcen Xu, Changrui Chen, Chunsheng Wu, Huajie Tan, Chunyuan Li, Jing Yang, Jie Yu, Xiyao Wang, Bin Qin, Yumeng Wang, Zizhen Yan, Ziyong Feng, Ziwei Liu, Bo Li, and Jiankang Deng. Llava-onevision-1.5: Fully open framework for democratized multimodal t...

    work page 2025

  4. [4]

    Qwen3-VL Technical Report

    [Baiet al., 2025a ] Shuai Bai, Yuxuan Cai, Ruizhe Chen, Ke- qin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhaohai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Junyang Lin, Xuejing Liu, Jiawei Liu, Chenglong Liu, Yang Liu, ...

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [5]

    Qwen2.5-VL Technical Report

    [Baiet al., 2025b ] Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. Qwen2.5-vl...

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    Meteor: An automatic metric for mt evaluation with improved correlation with human judgments

    [Banerjee and Lavie, 2005] Satanjeev Banerjee and Alon Lavie. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. InProceed- ings of the acl workshop on intrinsic and extrinsic evalua- tion measures for machine translation and/or summariza- tion, pages 65–72,

    work page 2005

  7. [7]

    Cascade r-cnn: Delving into high quality object detection

    [Cai and Vasconcelos, 2018] Zhaowei Cai and Nuno Vas- concelos. Cascade r-cnn: Delving into high quality object detection. InProceedings of the IEEE conference on com- puter vision and pattern recognition, pages 6154–6162,

    work page 2018

  8. [8]

    MMDetection: Open MMLab Detection Toolbox and Benchmark

    [Chenet al., 2019 ] Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, et al. Mmdetection: Open mmlab detection toolbox and benchmark.arXiv preprint arXiv:1906.07155,

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  9. [9]

    Lightweight network-based semantic segmentation for uavs and its risc-v implementation.Journal of Tech- nology Innovation and Engineering, 1(2),

    [Chenet al., 2025 ] Yankai Chen, Hongkun Du, and Yutong Zhou. Lightweight network-based semantic segmentation for uavs and its risc-v implementation.Journal of Tech- nology Innovation and Engineering, 1(2),

    work page 2025

  10. [10]

    A lightweight cnn model for uav-based image classification

    [Denget al., 2025 ] Xinjie Deng, Michael Shi, Burhan Khan, Yit Hong Choo, Fazal Ghaffar, and Chee Peng Lim. A lightweight cnn model for uav-based image classification. Soft Computing, 29(4):2363–2378,

    work page 2025

  11. [11]

    Visdrone-det2019: The vision meets drone object detection in image chal- lenge results

    [Duet al., 2019 ] Dawei Du, Pengfei Zhu, Longyin Wen, Xiao Bian, Haibin Lin, Qinghua Hu, Tao Peng, Jiayu Zheng, Xinyao Wang, Yue Zhang, et al. Visdrone-det2019: The vision meets drone object detection in image chal- lenge results. InProceedings of the IEEE/CVF interna- tional conference on computer vision workshops, pages 0– 0,

    work page 2019

  12. [12]

    YOLOX: Exceeding YOLO Series in 2021

    [Geet al., 2021 ] Zheng Ge, Songtao Liu, Feng Wang, Zem- ing Li, and Jian Sun. Yolox: Exceeding yolo series in 2021.arXiv preprint arXiv:2107.08430,

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  13. [13]

    Expanding open-vocabulary understanding for uav aerial imagery: A vision–language framework to semantic seg- mentation.Drones, 9(2):155,

    [Huanget al., 2025 ] Bangju Huang, Junhui Li, Wuyang Luan, Jintao Tan, Chenglong Li, and Longyang Huang. Expanding open-vocabulary understanding for uav aerial imagery: A vision–language framework to semantic seg- mentation.Drones, 9(2):155,

    work page 2025

  14. [14]

    A review: Obstacle percep- tion based on panoramic vision in low-altitude rotorcraft uavs.Applications of Modelling and Simulation, 9:213– 226,

    [Jianget al., 2025 ] Xiaoyan Jiang, Khairul Hamimah Abas, and Abdul Rashid Husain. A review: Obstacle percep- tion based on panoramic vision in low-altitude rotorcraft uavs.Applications of Modelling and Simulation, 9:213– 226,

    work page 2025

  15. [15]

    Rouge: A package for automatic evaluation of summaries

    [Lin, 2004] Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81,

    work page 2004

  16. [16]

    A multi-scale feature- fusion multi-object tracking algorithm for scale-variant ve- hicle tracking in uav videos.Remote Sensing, 17(6):1014,

    [Liuet al., 2025 ] Shanshan Liu, Xinglin Shen, Shanzhu Xiao, Hanwen Li, and Huamin Tao. A multi-scale feature- fusion multi-object tracking algorithm for scale-variant ve- hicle tracking in uav videos.Remote Sensing, 17(6):1014,

    work page 2025

  17. [17]

    Bleu: a method for au- tomatic evaluation of machine translation

    [Papineniet al., 2002 ] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for au- tomatic evaluation of machine translation. InProceedings of the 40th annual meeting of the Association for Compu- tational Linguistics, pages 311–318,

    work page 2002

  18. [18]

    Pytorch: An imperative style, high- performance deep learning library.Advances in neural in- formation processing systems, 32,

    [Paszkeet al., 2019 ] Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high- performance deep learning library.Advances in neural in- formation processing systems, 32,

    work page 2019

  19. [19]

    Faster r-cnn: Towards real-time object detection with region proposal networks.IEEE transactions on pattern analysis and machine intelligence, 39(6):1137–1149,

    [Renet al., 2016 ] Shaoqing Ren, Kaiming He, Ross Gir- shick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks.IEEE transactions on pattern analysis and machine intelligence, 39(6):1137–1149,

    work page 2016

  20. [20]

    Rareplanes: Synthetic data takes flight

    [Shermeyeret al., 2021 ] Jacob Shermeyer, Thomas Hossler, Adam Van Etten, Daniel Hogan, Ryan Lewis, and Daeil Kim. Rareplanes: Synthetic data takes flight. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 207–217,

    work page 2021

  21. [21]

    Drone-based rgb-infrared cross-modality vehicle detection via uncertainty-aware learning.IEEE Transactions on Circuits and Systems for Video Technol- ogy, 32(10):6700–6713,

    [Sunet al., 2022 ] Yiming Sun, Bing Cao, Pengfei Zhu, and Qinghua Hu. Drone-based rgb-infrared cross-modality vehicle detection via uncertainty-aware learning.IEEE Transactions on Circuits and Systems for Video Technol- ogy, 32(10):6700–6713,

    work page 2022

  22. [22]

    Cider: Consensus- based image description evaluation

    [Vedantamet al., 2015 ] Ramakrishna Vedantam, C Lawrence Zitnick, and Devi Parikh. Cider: Consensus- based image description evaluation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 4566–4575,

    work page 2015

  23. [23]

    InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency

    [Wanget al., 2025a ] Weiyun Wang, Zhangwei Gao, Lixin Gu, Hengjun Pu, Long Cui, Xingguang Wei, Zhaoyang Liu, Linglin Jing, Shenglong Ye, Jie Shao, et al. In- ternvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency.arXiv preprint arXiv:2508.18265,

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    Technological integration in urban emergency management: Challenges, innovations, and pathways to resilient systems.Science, Engineering and Technology Proceedings, 1:30–39,

    [Wang, 2025] Ziming Wang. Technological integration in urban emergency management: Challenges, innovations, and pathways to resilient systems.Science, Engineering and Technology Proceedings, 1:30–39,

    work page 2025

  25. [25]

    Transformers: State-of-the-art natural lan- guage processing

    [Wolfet al., 2020 ] Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Fun- towicz, et al. Transformers: State-of-the-art natural lan- guage processing. InProceedings of the 2020 conference on empirical methods in natural language processing: sys- tem demonstrations, pages 38–45,

    work page 2020

  26. [26]

    Qwen3 Technical Report

    [Yanget al., 2025a ] An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jingren Zhou, Junyang Lin, Kai Dang, Ke...

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    Mm-tracker: Motion mamba for uav- platform multiple object tracking

    [Yaoet al., 2025 ] Mufeng Yao, Jinlong Peng, Qingdong He, Bo Peng, Hao Chen, Mingmin Chi, Chao Liu, and Jon Atli Benediktsson. Mm-tracker: Motion mamba for uav- platform multiple object tracking. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 9409–9417,

    work page 2025

  28. [28]

    Scale match for tiny person detection

    [Yuet al., 2020 ] Xuehui Yu, Yuqi Gong, Nan Jiang, Qixi- ang Ye, and Zhenjun Han. Scale match for tiny person detection. InProceedings of the IEEE/CVF winter con- ference on applications of computer vision, pages 1257– 1265,

    work page 2020

  29. [29]

    Towards a low-altitude aerial in- telligent network: Vision, challenges, and key technolo- gies.China Communications, 22(9):1–21,

    [Yuhonget al., 2025 ] Huang Yuhong, Ding Haiyu, Chen Weiyan, Kong Luting, Deng Wei, Li Xin, Liu Yang, Wang Guizhen, and Liu Liang. Towards a low-altitude aerial in- telligent network: Vision, challenges, and key technolo- gies.China Communications, 22(9):1–21,

    work page 2025

  30. [30]

    DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection

    [Zhanget al., 2022 ] Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M Ni, and Heung- Yeung Shum. Dino: Detr with improved denoising an- chor boxes for end-to-end object detection.arXiv preprint arXiv:2203.03605,

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  31. [31]

    [Zhanget al., 2025 ] Zhen Zhang, Linhuan Jiang, Bo-Hui Tang, Jianchen Liu, Qingwang Wang, Yabin Hu, Liang Huang, and Zhitao Fu. Attention residual hybrid network for unmanned aerial vehicles hyperspectral image classifi- cation.IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,

    work page 2025

  32. [32]

    An open and comprehensive pipeline for unified object grounding and detection.arXiv preprint arXiv:2401.02361,

    [Zhaoet al., 2024 ] Xiangyu Zhao, Yicheng Chen, Shilin Xu, Xiangtai Li, Xinjiang Wang, Yining Li, and Ha- ian Huang. An open and comprehensive pipeline for unified object grounding and detection.arXiv preprint arXiv:2401.02361,

    work page arXiv 2024

  33. [33]

    Uav-detr: an enhanced rt-detr architecture for efficient small object de- tection in uav imagery.Sensors, 25(15):4582,

    [Zhou and Wei, 2025] Yu Zhou and Yan Wei. Uav-detr: an enhanced rt-detr architecture for efficient small object de- tection in uav imagery.Sensors, 25(15):4582,

    work page 2025

  34. [34]

    Deformable DETR: Deformable Transformers for End-to-End Object Detection

    [Zhuet al., 2020 ] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159,

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  35. [35]

    InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

    [Zhuet al., 2025 ] Jinguo Zhu, Weiyun Wang, Zhe Chen, Zhaoyang Liu, Shenglong Ye, Lixin Gu, Hao Tian, Yuchen Duan, Weijie Su, Jie Shao, et al. Internvl3: Exploring ad- vanced training and test-time recipes for open-source mul- timodal models.arXiv preprint arXiv:2504.10479, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025