pith. machine review for the scientific record. sign in

arxiv: 2604.15065 · v1 · submitted 2026-04-16 · 💻 cs.CV

Recognition: unknown

Learning Where to Embed: Noise-Aware Positional Embedding for Query Retrieval in Small-Object Detection

Yangchen Zeng , Zhenyu Yu , Dongming Jiang , Wenbo Zhang , Yifan Hong , Zhanhua Hu , Jiao Luo , Kangning Cui
Authors on Pith no claims yet

Pith reviewed 2026-05-10 12:07 UTC · model grok-4.3

classification 💻 cs.CV
keywords small-object detectionpositional embeddingtransformer decoderquery noise suppressionheatmap guidanceefficient object detectionbackground clutterquery retrieval
0
0 comments X

The pith

Noise-aware positional embedding enables shallower decoders for accurate small-object detection in transformers.

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

Transformer-based detectors for small objects suffer from background-induced noise in their queries, often requiring deep decoder stacks to refine them. This work proposes a framework that learns to embed positional information selectively in foreground regions using heatmap guidance. By suppressing background clutter in the embeddings, it retrieves higher-quality queries earlier in the process. Consequently, the decoder can be reduced from eight layers to three while cutting parameters by nearly 60 percent and still achieving accuracy improvements on benchmarks. This efficiency gain matters for deploying robust detection in constrained environments where compute is limited.

Core claim

The central claim is that a Heatmap-guided Embedding Learning Paradigm (HELP) with Heatmap-guided Positional Embedding (HPE) can guide feature encoding and query filtering to suppress background noise, allowing high-quality small-object query retrieval with significantly reduced decoder depth and parameters, all without extra inference cost.

What carries the argument

Heatmap-guided Positional Embedding (HPE), a mechanism that injects positional encodings selectively based on heatmap salience to preserve foreground information and filter out background-dominant queries via a gradient-based mask.

If this is right

  • Decoder layers reduced from eight to three while maintaining accuracy gains.
  • Model parameters reduced by 59.4% from 163M to 66.3M.
  • Consistent performance improvements across benchmarks under lower compute budgets.
  • Linear-Snake Convolution enriches sparse features for better small-target representation.
  • Gradient supervision for the mask applies only at training time, with zero added cost at inference.

Where Pith is reading between the lines

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

  • This approach of selective positional embedding might generalize to other attention-based vision tasks where background noise affects query quality.
  • Combining this noise suppression with other efficiency methods like knowledge distillation could yield even more compact detectors for mobile applications.
  • Evaluating the method on additional datasets with varying object densities could confirm its robustness beyond the reported benchmarks.

Load-bearing premise

The heatmap guidance and mask filter can reliably separate background noise from useful positional signals for small objects without removing essential localization information.

What would settle it

Running the three-layer decoder version on a standard small-object detection benchmark and observing whether its average precision falls below that of the original eight-layer model without the proposed embedding.

Figures

Figures reproduced from arXiv: 2604.15065 by Dongming Jiang, Jiao Luo, Kangning Cui, Wenbo Zhang, Yangchen Zeng, Yifan Hong, Zhanhua Hu, Zhenyu Yu.

Figure 1
Figure 1. Figure 1: Overview of HELP. Backbone: LSConv is integrated into the backbone to enhance sparse features before [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: HPE shows where to embed. Top: the encoder fuses heatmap information into embeddings. Middle: the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Heatmap-guided query retrieval in an encoder-decoder detector. HPE produces a binary mask from heatmaps [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Dual-path axis-wise sampling in LSConv. Left: x-/y-axis convolution samples on a [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sampling patterns comparison on a 9 × 9 grid. We contrast standard, dilated, deformable [53], and DSC [54] convolutions with two LSConv variants. LSConv combines a constrained linear path and a deformable snake-like path to cover both straight and curved structures, while other operators follow their fixed or unconstrained sampling layouts. preventing excessive deformation that could harm alignment. For cl… view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of HPE heatbars under cluttered versus clean scenes. The heatmap guidance provides more [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
read the original abstract

Transformer-based detectors have advanced small-object detection, but they often remain inefficient and vulnerable to background-induced query noise, which motivates deep decoders to refine low-quality queries. We present HELP (Heatmap-guided Embedding Learning Paradigm), a noise-aware positional-semantic fusion framework that studies where to embed positional information by selectively preserving positional encodings in foreground-salient regions while suppressing background clutter. Within HELP, we introduce Heatmap-guided Positional Embedding (HPE) as the core embedding mechanism and visualize it with a heatbar for interpretable diagnosis and fine-tuning. HPE is integrated into both the encoder and decoder: it guides noise-suppressed feature encoding by injecting heatmap-aware positional encoding, and it enables high-quality query retrieval by filtering background-dominant embeddings via a gradient-based mask filter before decoding. To address feature sparsity in complex small targets, we integrate Linear-Snake Convolution to enrich retrieval-relevant representations. The gradient-based heatmap supervision is used during training only, incurring no additional gradient computation at inference. As a result, our design reduces decoder layers from eight to three and achieves a 59.4% parameter reduction (66.3M vs. 163M) while maintaining consistent accuracy gains under a reduced compute budget across benchmarks. Code Repository: https://github.com/yidimopozhibai/Noise-Suppressed-Query-Retrieval

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 HELP (Heatmap-guided Embedding Learning Paradigm), a noise-aware positional-semantic fusion framework for transformer-based small-object detection. It proposes Heatmap-guided Positional Embedding (HPE) to selectively preserve positional encodings in foreground regions while suppressing background clutter via a gradient-based mask filter, integrated into both encoder and decoder; Linear-Snake Convolution is added to address feature sparsity. The central empirical claim is that this enables reducing decoder layers from 8 to 3, yielding a 59.4% parameter reduction (66.3M vs. 163M) while maintaining consistent accuracy gains across benchmarks under lower compute, with heatmap supervision used only at training time.

Significance. If the accuracy claims hold after rigorous validation, the work could meaningfully advance efficient transformer detectors for small objects by mitigating query noise without deep decoders, which is relevant for resource-constrained applications such as aerial imagery or medical imaging. The interpretable heatbar visualization for HPE and the training-only supervision are practical strengths; the code repository is a positive for reproducibility.

major comments (2)
  1. [Abstract and §3.2] Abstract and §3.2 (gradient-based mask filter): The headline result of maintaining accuracy with a 3-layer decoder and 59.4% parameter reduction rests on the untested premise that the mask reliably separates foreground-salient positional encodings from background clutter at small-object scales without discarding localization cues. No ablation isolating the mask's contribution (e.g., vs. Linear-Snake Convolution alone) or quantitative metrics on false-negative rates for small objects is referenced, which directly undermines the reduced-decoder claim.
  2. [§4] §4 (experiments): The abstract states 'consistent accuracy gains' and efficiency numbers, but the provided details lack full baseline comparisons, error bars, or cross-benchmark tables showing that gains persist when decoder depth is fixed and only HPE+mask are added. This makes it impossible to confirm the data support the central efficiency-accuracy tradeoff.
minor comments (2)
  1. [§3] The new terms 'Heatmap-guided Positional Embedding (HPE)' and 'Linear-Snake Convolution' are introduced without explicit equations or pseudocode in the abstract; adding these in §3 would improve clarity and allow direct comparison to prior positional embeddings (e.g., in DETR variants).
  2. [Figures] The heatbar visualization is mentioned for interpretable diagnosis; ensure all figures include scale bars and are referenced in the text for fine-tuning guidance.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive review. The comments correctly identify areas where additional ablations and experimental details would strengthen the presentation of the efficiency-accuracy claims. We address each point below and will incorporate the requested analyses in the revised manuscript.

read point-by-point responses

  1. Referee: [Abstract and §3.2] Abstract and §3.2 (gradient-based mask filter): The headline result of maintaining accuracy with a 3-layer decoder and 59.4% parameter reduction rests on the untested premise that the mask reliably separates foreground-salient positional encodings from background clutter at small-object scales without discarding localization cues. No ablation isolating the mask's contribution (e.g., vs. Linear-Snake Convolution alone) or quantitative metrics on false-negative rates for small objects is referenced, which directly undermines the reduced-decoder claim.

    Authors: We agree that an explicit ablation isolating the gradient-based mask filter is necessary to substantiate the reduced-decoder claim. In the revision we will add a dedicated ablation table comparing (i) the full HELP model, (ii) HPE without the mask, (iii) Linear-Snake Convolution alone, and (iv) the mask applied without HPE. We will also report quantitative false-negative rates on small-object localization (measured via bounding-box IoU thresholds) to quantify any loss of localization cues. These additions will directly address the concern while preserving the training-only nature of the supervision. revision: yes

  2. Referee: [§4] §4 (experiments): The abstract states 'consistent accuracy gains' and efficiency numbers, but the provided details lack full baseline comparisons, error bars, or cross-benchmark tables showing that gains persist when decoder depth is fixed and only HPE+mask are added. This makes it impossible to confirm the data support the central efficiency-accuracy tradeoff.

    Authors: We acknowledge that the current experimental section would benefit from more granular controls. In the revision we will expand §4 with: (a) full baseline tables including standard DETR and Deformable DETR at decoder depths 8, 6, 4, and 3; (b) results with error bars from three independent runs; and (c) cross-benchmark tables that fix decoder depth at 3 and incrementally add only HPE and the mask. These controlled experiments will isolate the contribution of the proposed components to the observed efficiency-accuracy tradeoff. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical architectural design with no derivational reduction

full rationale

The manuscript introduces HELP as an empirical noise-aware embedding framework for small-object detection, relying on proposed components (HPE, gradient-based mask filter, Linear-Snake Convolution) whose benefits are asserted via benchmark experiments rather than any closed mathematical derivation. No equations, fitted parameters renamed as predictions, or self-citation chains appear in the load-bearing claims; the decoder-layer reduction and parameter savings are presented as observed outcomes of the design choices, not as quantities forced by construction from the inputs. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Based on the abstract, the central claim rests on the effectiveness of newly introduced components whose internal hyperparameters and exact integration details are not specified.

invented entities (2)
  • Heatmap-guided Positional Embedding (HPE) no independent evidence
    purpose: Selectively preserving positional encodings in foreground-salient regions while suppressing background clutter
    Core new embedding mechanism introduced by the paper.
  • Linear-Snake Convolution no independent evidence
    purpose: Enriching retrieval-relevant representations for sparse small-object features
    New convolution variant proposed to address feature sparsity.

pith-pipeline@v0.9.0 · 5565 in / 1336 out tokens · 56084 ms · 2026-05-10T12:07:15.792861+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Mind the Unseen Mass: Unmasking LLM Hallucinations via Soft-Hybrid Alphabet Estimation

    cs.CL 2026-04 unverdicted novelty 6.0

    SHADE adaptively combines coverage and spectral signals to estimate semantic alphabet size from few LLM samples, yielding better performance than baselines in low-sample regimes for alphabet estimation and QA error detection.

Reference graph

Works this paper leans on

73 extracted references · 13 canonical work pages · cited by 1 Pith paper · 5 internal anchors

  1. [1]

    A review of small object detection based on deep learning

    Wei Wei, Yu Cheng, Jiafeng He, and Xiyue Zhu. A review of small object detection based on deep learning. Neural Comput. Appl., 36(12):6283–6303, 2024

    2024

  2. [2]

    Scrdet++: Detecting small, cluttered and rotated objects via instance-level feature denoising and rotation loss smoothing.IEEE TPAMI, 2022

    Xue Yang, Junchi Yan, Wenlong Liao, Xiaokang Yang, Jin Tang, and Tao He. Scrdet++: Detecting small, cluttered and rotated objects via instance-level feature denoising and rotation loss smoothing.IEEE TPAMI, 2022

    2022

  3. [3]

    Efficient unsupervised clustering of hyperspectral images via flexible multi-anchor graphs.Remote Sens., 17(15):2647, 2025

    Yihong Li, Ting Wang, Zhe Cao, Haonan Xin, and Rong Wang. Efficient unsupervised clustering of hyperspectral images via flexible multi-anchor graphs.Remote Sens., 17(15):2647, 2025

    2025

  4. [4]

    Mdpcaps-csl for sar image target recognition with limited labeled training data.IEEE Access, 8:176217–176231, 2020

    Yuchao Hou, Ting Xu, Hongping Hu, Peng Wang, Hongxin Xue, and Yanping Bai. Mdpcaps-csl for sar image target recognition with limited labeled training data.IEEE Access, 8:176217–176231, 2020

    2020

  5. [5]

    Palmprobnet: A probabilistic approach to understanding palm distributions in ecuadorian tropical forest via transfer learning

    Kangning Cui, Zishan Shao, Gregory Larsen, Victor Pauca, Sarra Alqahtani, David Segurado, João Pinheiro, Manqi Wang, David Lutz, Robert Plemmons, et al. Palmprobnet: A probabilistic approach to understanding palm distributions in ecuadorian tropical forest via transfer learning. InProceedings of the 2024 ACM Southeast Conference, pages 272–277, 2024

    2024

  6. [6]

    Superpixel-based and spatially regularized diffusion learning for unsupervised hyperspectral image clustering.IEEE Transactions on Geoscience and Remote Sensing, 62:1–18, 2024

    Kangning Cui, Ruoning Li, Sam L Polk, Yinyi Lin, Hongsheng Zhang, James M Murphy, Robert J Plemmons, and Raymond H Chan. Superpixel-based and spatially regularized diffusion learning for unsupervised hyperspectral image clustering.IEEE Transactions on Geoscience and Remote Sensing, 62:1–18, 2024

    2024

  7. [7]

    Towards large-scale small object detection: Survey and benchmarks.IEEE TPAMI, 45(11):13467–13488, 2023

    Gong Cheng, Xiang Yuan, Xiwen Yao, Kebing Yan, Qinghua Zeng, Xingxing Xie, and Junwei Han. Towards large-scale small object detection: Survey and benchmarks.IEEE TPAMI, 45(11):13467–13488, 2023

    2023

  8. [8]

    Efficient localization and spatial distribution modeling of canopy palms using uav imagery.IEEE TGRS, 2025

    Kangning Cui, Wei Tang, Rongkun Zhu, Manqi Wang, Gregory D Larsen, Victor P Pauca, Sarra Alqahtani, Fan Yang, David Segurado, Paul Fine, et al. Efficient localization and spatial distribution modeling of canopy palms using uav imagery.IEEE TGRS, 2025

    2025

  9. [9]

    Slicing aided hyper inference and fine-tuning for small object detection

    Fatih Cagatay Akyon, Sinan Onur Altinuc, and Alptekin Temizel. Slicing aided hyper inference and fine-tuning for small object detection. InICIP, pages 966–970. IEEE, 2022

    2022

  10. [10]

    MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents

    Dongming Jiang, Yi Li, Guanpeng Li, and Bingzhe Li. Magma: A multi-graph based agentic memory architecture for ai agents.arXiv preprint arXiv:2601.03236, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  11. [11]

    Detrs beat yolos on real-time object detection

    Yian Zhao, Wenyu Lv, Shangliang Xu, Jinman Wei, Guanzhong Wang, Qingqing Dang, Yi Liu, and Jie Chen. Detrs beat yolos on real-time object detection. InCVPR, pages 16965–16974, 2024

    2024

  12. [12]

    Dino: Detr with improved denoising anchor boxes for end-to-end object detection

    Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel Ni, and Harry Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. InICLR, 2023

    2023

  13. [13]

    Dinov3 visual representations for blueberry perception toward robotic harvesting.arXiv preprint arXiv:2603.02419, 2026

    Rui-Feng Wang, Daniel Petti, Yue Chen, and Changying Li. Dinov3 visual representations for blueberry perception toward robotic harvesting.arXiv preprint arXiv:2603.02419, 2026

    work page arXiv 2026

  14. [14]

    From orthomosaics to raw uav imagery: Enhancing palm detection and crown-center localization.arXiv preprint arXiv:2509.12400, 2025

    Rongkun Zhu, Kangning Cui, Wei Tang, Rui-Feng Wang, Sarra Alqahtani, David Lutz, Fan Yang, Paul Fine, Jordan Karubian, Robert Plemmons, et al. From orthomosaics to raw uav imagery: Enhancing palm detection and crown-center localization.arXiv preprint arXiv:2509.12400, 2025. 11 Running Title for Header

    work page arXiv 2025

  15. [15]

    Conditional for fast training convergence

    Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, and Jingdong Wang. Conditional for fast training convergence. InICCV, pages 3651–3660, 2021

    2021

  16. [16]

    Dn-detr: Accelerate detr training by introducing query denoising

    Feng Li, Hao Zhang, Shilong Liu, Jian Guo, Lionel M Ni, and Lei Zhang. Dn-detr: Accelerate detr training by introducing query denoising. InCVPR, pages 13619–13627, 2022

    2022

  17. [17]

    arXiv preprint arXiv:2602.19320 , year=

    Dongming Jiang, Yi Li, Songtao Wei, Jinxin Yang, Ayushi Kishore, Alysa Zhao, Dingyi Kang, Xu Hu, Feng Chen, Qiannan Li, et al. Anatomy of agentic memory: Taxonomy and empirical analysis of evaluation and system limitations.arXiv preprint arXiv:2602.19320, 2026

    work page arXiv 2026

  18. [18]

    Center-guided classifier for semantic segmentation of remote sensing images.IEEE TGRS, 2026

    Wei Zhang, Qin Huang, Mengting Ma, Yizhen Jiang, Yun Chen, Zhenhua Huang, Wangyu Wu, Kangning Cui, Rongrong Lian, Zhenkai Wu, and Xiaowen Ma. Center-guided classifier for semantic segmentation of remote sensing images.IEEE TGRS, 2026

    2026

  19. [19]

    Implicit regression in subspace for high-sensitivity cest imaging

    Chu Chen, Yang Liu, Se Weon Park, Jizhou Li, Kannie WY Chan, and Raymond HF Chan. Implicit regression in subspace for high-sensitivity cest imaging. In2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5. IEEE, 2024

    2024

  20. [20]

    Blind adaptive local denoising for cest imaging.arXiv preprint arXiv:2511.20081, 2025

    Chu Chen, Aitor Artola, Yang Liu, Se Weon Park, Raymond H Chan, Jean-Michel Morel, and Kannie WY Chan. Blind adaptive local denoising for cest imaging.arXiv preprint arXiv:2511.20081, 2025

    work page arXiv 2025

  21. [21]

    Sada: Stability-guided adaptive diffusion acceleration

    Ting Jiang, Yixiao Wang, Hancheng Ye, Zishan Shao, Jingwei Sun, Jingyang Zhang, Zekai Chen, Jianyi Zhang, Yiran Chen, and Hai Li. Sada: Stability-guided adaptive diffusion acceleration. InForty-second International Conference on Machine Learning, 2025

    2025

  22. [22]

    Toward efficient uav-based small object detection: A lightweight network with enhanced feature fusion.Remote Sens., 17(13):2235, 2025

    Xingyu Di, Kangning Cui, and Rui-Feng Wang. Toward efficient uav-based small object detection: A lightweight network with enhanced feature fusion.Remote Sens., 17(13):2235, 2025

    2025

  23. [23]

    Seaf-net: A sustainable and lightweight attention-enhanced detection network for underwater fish species recognition.Journal of Marine Science and Engineering, 14(4):351, 2026

    Yu-Shan Han, Sheng-Lun Zhao, Chu Chen, Kangning Cui, Pingfan Hu, and Rui-Feng Wang. Seaf-net: A sustainable and lightweight attention-enhanced detection network for underwater fish species recognition.Journal of Marine Science and Engineering, 14(4):351, 2026

    2026

  24. [24]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016

    2016

  25. [25]

    Accurate detection and instance segmentation of unstained living adherent cells in differential interference contrast images.Computers in Biology and Medicine, 182:109151, 2024

    Fei Pan, Yutong Wu, Kangning Cui, Shuxun Chen, Yanfang Li, Yaofang Liu, Adnan Shakoor, Han Zhao, Beijia Lu, Shaohua Zhi, et al. Accurate detection and instance segmentation of unstained living adherent cells in differential interference contrast images.Computers in Biology and Medicine, 182:109151, 2024

    2024

  26. [26]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. InECCV, 2016

    2016

  27. [27]

    Yolov3: An incremental improvement

    Ali Farhadi, Joseph Redmon, et al. Yolov3: An incremental improvement. InCVPR, volume 1804, pages 1–6. Springer Berlin/Heidelberg, Germany, 2018

    2018

  28. [28]

    Behavior recognition and tracking of group-housed pigs based on improved bytetrack algorithm.Trans

    Shuqin Tu, Yunjie Tang, Chengjie Li, Yun Liang, Yangchen Zeng, and Xiaolong Liu. Behavior recognition and tracking of group-housed pigs based on improved bytetrack algorithm.Trans. Chin. Soc. Agric. Mach., 53(12):264–272, 2022

    2022

  29. [29]

    Bmdnet-yolo: A lightweight and robust model for high-precision real-time recognition of blueberry maturity.Horticulturae, 11(10):1202, 2025

    Huihui Sun and Rui-Feng Wang. Bmdnet-yolo: A lightweight and robust model for high-precision real-time recognition of blueberry maturity.Horticulturae, 11(10):1202, 2025

    2025

  30. [30]

    Cott-adnet: Lightweight real-time cotton boll and flower detection under field conditions

    Rui-Feng Wang, Mingrui Xu, Matthew Bauer, Iago Schardong, Xiaowen Ma, Peng Chee, and Kangning Cui. Cott-adnet: Lightweight real-time cotton boll and flower detection under field conditions. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 500–509, 2026

    2026

  31. [31]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In ICCV, 2017

    2017

  32. [32]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. InNeurIPS, 2015

    2015

  33. [33]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. Mask r-cnn. InICCV, 2017

    2017

  34. [34]

    Fedc-dac: A federated clustering with dynamic aggregation and calibration method for sar image target recognition.IEEE JSTAR, 2025

    Yuchao Hou, Shuai Zhao, Xiaoyu Xia, Minghui Liwang, Zijian Li, Nan Xu, Di Wu, Youliang Tian, and Tony QS Quek. Fedc-dac: A federated clustering with dynamic aggregation and calibration method for sar image target recognition.IEEE JSTAR, 2025

    2025

  35. [35]

    Luffd-yolo: A lightweight model for uav remote sensing forest fire detection based on attention mechanism and multi-level feature fusion.Remote Sens., 16(12):2177, 2024

    Yuhang Han, Bingchen Duan, Renxiang Guan, Guang Yang, and Zhen Zhen. Luffd-yolo: A lightweight model for uav remote sensing forest fire detection based on attention mechanism and multi-level feature fusion.Remote Sens., 16(12):2177, 2024. 12 Running Title for Header

    2024

  36. [36]

    Superyolo: Super resolution assisted object detection in multimodal remote sensing imagery.IEEE TGRS, 61:1–15, 2023

    Jiaqing Zhang, Jie Lei, Weiying Xie, Zhenman Fang, Yunsong Li, and Qian Du. Superyolo: Super resolution assisted object detection in multimodal remote sensing imagery.IEEE TGRS, 61:1–15, 2023

    2023

  37. [37]

    Pcnet: A structure similarity enhancement method for multispectral and multimodal image registration.Inf

    Si-Yuan Cao, Beinan Yu, Lun Luo, Runmin Zhang, Shu-Jie Chen, Chunguang Li, and Hui-Liang Shen. Pcnet: A structure similarity enhancement method for multispectral and multimodal image registration.Inf. Fusion, 94:200–214, 2023

    2023

  38. [38]

    Learning non-maximum suppression

    Jan Hosang, Rodrigo Benenson, and Bernt Schiele. Learning non-maximum suppression. InCVPR, pages 4507–4515, 2017

    2017

  39. [39]

    Cascade r-cnn: Delving into high quality object detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: Delving into high quality object detection. InCVPR, 2018

    2018

  40. [40]

    Larsen, Victor P

    Kangning Cui, Rongkun Zhu, Manqi Wang, Wei Tang, Gregory D. Larsen, Victor P. Pauca, Sarra Alqahtani, Fan Yang, David Segurado, David A. Lutz, Jean-Michel Morel, and Miles R. Silman. Detection and geographic localization of natural objects in the wild: A case study on palms. InIJCAI, pages 9601–9609. International Joint Conferences on Artificial Intellige...

    2025

  41. [41]

    End-to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. InECCV, pages 213–229. Springer, 2020

    2020

  42. [42]

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

    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, 2020

    work page internal anchor Pith review arXiv 2010

  43. [43]

    Dynamicvit: Efficient vision transformers with dynamic token sparsification

    Yongming Rao, Wenliang Zhao, Benlin Liu, Jiwen Lu, Jie Zhou, and Cho-Jui Hsieh. Dynamicvit: Efficient vision transformers with dynamic token sparsification. InNeurIPS, 2021

    2021

  44. [44]

    A coarse to fine detection method for prohibited object in x-ray images based on progressive transformer decoder

    Chunjie Ma, Lina Du, Zan Gao, Li Zhuo, and Meng Wang. A coarse to fine detection method for prohibited object in x-ray images based on progressive transformer decoder. InACM MM, pages 2700–2708, 2024

    2024

  45. [45]

    Sparseformer: Detecting objects in hrw shots via sparse vision transformer

    Wenxi Li, Yuchen Guo, Jilai Zheng, Haozhe Lin, Chao Ma, Lu Fang, and Xiaokang Yang. Sparseformer: Detecting objects in hrw shots via sparse vision transformer. InACM MM, pages 4851–4860, 2024

    2024

  46. [46]

    Optimized hard exudate detection with supervised contrastive learning

    Wei Tang, Kangning Cui, and Raymond H Chan. Optimized hard exudate detection with supervised contrastive learning. In2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5. IEEE, 2024

    2024

  47. [47]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. InICCV, 2021

    2021

  48. [48]

    Pyramid vision transformer: A versatile backbone for dense prediction without convolutions

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. InICCV, 2021

    2021

  49. [49]

    Cvt: Introducing convolutions to vision transformers

    Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt: Introducing convolutions to vision transformers. InICCV, 2021

    2021

  50. [50]

    Nyströmformer: A nyström-based algorithm for approximating self-attention

    Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, and Vikas Singh. Nyströmformer: A nyström-based algorithm for approximating self-attention. InAAAI, 2021

    2021

  51. [51]

    Flashsvd: Memory-efficient inference with streaming for low-rank models

    Zishan Shao, Yixiao Wang, Qinsi Wang, Ting Jiang, Zhixu Du, Hancheng Ye, Danyang Zhuo, Yiran Chen, et al. Flashsvd: Memory-efficient inference with streaming for low-rank models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 25278–25285, 2026

    2026

  52. [52]

    Query pipeline optimization for cancer patient question answering systems

    Maolin He, Rena Gao, Mike Conway, and Brian E Chapman. Query pipeline optimization for cancer patient question answering systems.arXiv preprint arXiv:2412.14751, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  53. [53]

    Deformable convnets v2: More deformable, better results

    Xizhou Zhu, Han Hu, Stephen Lin, and Jifeng Dai. Deformable convnets v2: More deformable, better results. In CVPR, pages 9308–9316, 2019

    2019

  54. [54]

    Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation

    Yaolei Qi, Yuting He, Xiaoming Qi, Yuan Zhang, and Guanyu Yang. Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation. InICCV, pages 6070–6079, 2023

    2023

  55. [55]

    Learning rotation-invariant convolutional neural networks for object detection in vhr optical remote sensing images.IEEE TGRS, 54(12):7405–7415, 2016

    Gong Cheng, Peicheng Zhou, and Junwei Han. Learning rotation-invariant convolutional neural networks for object detection in vhr optical remote sensing images.IEEE TGRS, 54(12):7405–7415, 2016

    2016

  56. [56]

    The pascal visual object classes (voc) challenge.IJCV, 2010

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge.IJCV, 2010

    2010

  57. [57]

    Visualizing our changing earth: A creative ai framework for democratizing environmental storytelling through satellite imagery

    Zhenyu Yu, Mohd Yamani Idna Idris, and Pei Wang. Visualizing our changing earth: A creative ai framework for democratizing environmental storytelling through satellite imagery. InNeurIPS 2025, 2025

    2025

  58. [58]

    Iidm: Improved implicit diffusion model with knowledge distillation to estimate the spatial distribution density of carbon stock in remote sensing imagery.KBS, page 115131, 2025

    Zhenyu Yu, Jinnian Wang, and Mohd Yamani Idna Idris. Iidm: Improved implicit diffusion model with knowledge distillation to estimate the spatial distribution density of carbon stock in remote sensing imagery.KBS, page 115131, 2025. 13 Running Title for Header

    2025

  59. [59]

    Dinov3-powered multi-task foundation model for quantitative remote sensing estimation.AAAI 2026, 40(48):41455–41456, 2026

    Zhenyu Yu, Mohd Yamani Idna Idris, Pei Wang, and Rizwan Qureshi. Dinov3-powered multi-task foundation model for quantitative remote sensing estimation.AAAI 2026, 40(48):41455–41456, 2026

    2026

  60. [60]

    Spatiotemporal alignment for remote sensing image recovery via terrain-aware diffusion.ICASSP 2026, 2026

    Zhenyu Yu, Haoran Jiang, Pei Wang, Zizhen Lin, and Yong Xiang. Spatiotemporal alignment for remote sensing image recovery via terrain-aware diffusion.ICASSP 2026, 2026

    2026

  61. [61]

    Fcos: Fully convolutional one-stage object detection

    Zhi Tian, Chunhua Shen, Hao Chen, and Tong He. Fcos: Fully convolutional one-stage object detection. InICCV, pages 9627–9636, 2019

    2019

  62. [62]

    Centernet: Keypoint triplets for object detection

    Kaiwen Duan, Song Bai, Lingxi Xie, Honggang Qi, Qingming Huang, and Qi Tian. Centernet: Keypoint triplets for object detection. InICCV, pages 6569–6578, 2019

    2019

  63. [63]

    Searching for mobilenetv3

    Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. InICCV, pages 1314–1324, 2019

    2019

  64. [64]

    YOLOv4: Optimal Speed and Accuracy of Object Detection

    Alexey Bochkovskiy, Chien-Yao Wang, and Hong-Yuan Mark Liao. Yolov4: Optimal speed and accuracy of object detection.arXiv preprint arXiv:2004.10934, 2020

    work page internal anchor Pith review arXiv 2004

  65. [65]

    Yolo-z: Improving small object detection in yolov5 for autonomous vehicles.arXiv preprint arXiv:2112.11798, 2021

    Aduen Benjumea, Izzeddin Teeti, Fabio Cuzzolin, and Andrew Bradley. Yolo-z: Improving small object detection in yolov5 for autonomous vehicles.arXiv preprint arXiv:2112.11798, 2021

    work page arXiv 2021

  66. [66]

    Yolov6: A single-stage object detection framework for industrial applications

    Chuyi Li, Lulu Li, Hongliang Jiang, Kaiheng Weng, Yifei Geng, Liang Li, Zaidan Ke, Qingyuan Li, Meng Cheng, Weiqiang Nie, et al. Yolov6: A single-stage object detection framework for industrial applications.arXiv preprint arXiv:2209.02976, 2022

    work page arXiv 2022

  67. [67]

    Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors

    Chien-Yao Wang, Alexey Bochkovskiy, and Hong-Yuan Mark Liao. Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. InCVPR, pages 7464–7475, 2023

    2023

  68. [68]

    YOLOv11: An Overview of the Key Architectural Enhancements

    Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architectural enhancements.arXiv preprint arXiv:2410.17725, 2024

    work page internal anchor Pith review arXiv 2024

  69. [69]

    The footprint of factorization models and their applications in collaborative filtering.ACM Transactions on Information Systems (TOIS), 40(4):1–32, 2021

    Jinze Wang, Yongli Ren, Jie Li, and Ke Deng. The footprint of factorization models and their applications in collaborative filtering.ACM Transactions on Information Systems (TOIS), 40(4):1–32, 2021

    2021

  70. [70]

    Ads-poi: Agentic spatiotemporal state decomposition for next point-of-interest recommendation.ACM ISBN, pages 978–1, 2026

    Zhenyu Yu, Chunlei Meng, Yangchen Zeng, Mohd Yamani Idna Idris, and Shuigeng Zhou. Ads-poi: Agentic spatiotemporal state decomposition for next point-of-interest recommendation.ACM ISBN, pages 978–1, 2026

    2026

  71. [71]

    Cast-poi: Candidate- conditioned spatiotemporal modeling for next poi recommendation.ACM ISBN, pages 978–1, 2026

    Zhenyu Yu, Chunlei Meng, Yangchen Zeng, Mohd Yamani Idna Idris, and Shuigeng Zhou. Cast-poi: Candidate- conditioned spatiotemporal modeling for next poi recommendation.ACM ISBN, pages 978–1, 2026

    2026

  72. [72]

    Hmpe: Heatmap embedding for efficient transformer-based small object detection.arXiv preprint arXiv:2504.13469, 2025

    YangChen Zeng. Hmpe: Heatmap embedding for efficient transformer-based small object detection.arXiv preprint arXiv:2504.13469, 2025

    work page arXiv 2025

  73. [73]

    Deepinterestgr: Mining deep multi-interest using multi-modal llms for generative recommenda- tion.arXiv preprint arXiv:2602.18907, 2026

    Yangchen Zeng. Deepinterestgr: Mining deep multi-interest using multi-modal llms for generative recommenda- tion.arXiv preprint arXiv:2602.18907, 2026. 14

    work page arXiv 2026