CSDN: A Context-Gated Self-Adaptive Detection Network for Real-Time Object Detection

Haolin Wei

REVIEW 5 major objections 5 minor 28 references

CSDN: A Context-Gated Self-Adaptive Detection Network for Real-Time Object Detection

T0 review · 5 major / 5 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 A gated fusion of block, neighbor, and deformable attention, used as a drop-in detection head, lifts YOLO models by up to 1.0 AP on COCO after only 24 epochs of fine-tuning.

desk verdict A cleanly written incremental detection-head paper whose +0.7–1.0 AP claim is real only if the gains survive a controlled comparison against the same amount of fine-tuning. read the letter →

arxiv 2506.17679 v2 pith:BT2NB2CQ submitted 2025-06-21 cs.CV

Haolin Wei This is my paper

classification cs.CV

keywords CSDNobjectdetectionheadgatedfusionblockattentionneighbordeformablereal-time

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

The paper tries to show that a Transformer-based detection head called CSDN can be dropped into existing CNN detectors, specifically the YOLO family, and improve accuracy with very little retraining. The head replaces the stacked self-attention and cross-attention layers of DETR-style detectors with three specialized attention patterns that mimic human observation: a global scene summary, local neighbor context, and fine-grained deformable detail. A gating network weights these three patterns for each region of interest, making attention adaptive per object rather than uniform across all queries. If the claim holds, any YOLO-class detector can gain 0.7 to 1.0 AP on COCO while keeping real-time latency around 23 ms, without changing the backbone.

What carries the argument

The load-bearing component is a gated fusion of three attention modules. Block Attention cross-attends each query to the top-level FPN feature map for a compact global scene summary. Neighbor Attention restricts self-attention to queries whose bounding boxes overlap (IoU greater than zero) for local context. Deformable Attention predicts learned sampling offsets to extract fine details at key locations. A small gating network computes per-ROI weights that combine the three outputs, replacing a fixed stack of self- and cross-attention layers with an adaptive, scale-aware selection mechanism.

What would settle it

Re-run YOLOv5x, YOLOv8x, and YOLOv11x with their native heads under the exact 24-epoch RT-DETR-style training recipe used for CSDN and compare AP; if any native head reaches or exceeds the CSDN number (for example, YOLOv8x at 55.1 AP), the reported swap-in gain is not reproduced.

Watch

Extended reading notes

Core claim

CSDN's central claim is that exhaustive global self-attention among object queries in DETR-like heads is largely redundant, and that replacing it with three complementary, gated attention patterns improves both accuracy and efficiency. Concretely, the paper reports +1.0 AP on YOLOv5x (53.4 to 54.4) and YOLOv8x (54.1 to 55.1), and +0.7 AP on YOLOv11x (54.9 to 55.6), after only 24 epochs of fine-tuning on pre-trained weights. Ablations on YOLOv8 show that gated fusion of block, neighbor, and deformable attention reaches 71.5 mAP50 and 55.1 mAP50-95, beating the stacked self-attention baseline of 70.3 and 54.5. The paper also finds diminishing returns beyond four transformer layers, with four layers matching six layers in accuracy at lower latency.

Load-bearing premise

The reported gains assume that the published AP scores for YOLOv5x, YOLOv8x, and YOLOv11x are directly comparable to the authors' own 24-epoch fine-tuning runs; if those baselines improve under identical re-training or run-to-run variance exceeds about 0.5 AP, the +0.7 to +1.0 gain may disappear, a risk that enters in Section 4.2 (Table 3) and Section 4.3 (Table 5).

Editorial extensions

If this is right

Any CNN detector with a YOLO-style backbone can upgrade its head without retraining the backbone, needing only a short fine-tune of about 24 epochs.
The redundant global self-attention in DETR-like heads can be replaced by a small set of specialized attention patterns with no accuracy loss, and often with gains.
Real-time detection systems can keep latency near 23 ms while gaining accuracy, narrowing the gap between YOLO-class detectors and heavier DETR variants.
Layer depth shows diminishing returns beyond four transformer layers, so practitioners can trade one layer for lower latency while keeping accuracy.
The plug-and-play nature of the head means it can be applied across evolving YOLO architectures, as demonstrated on three successive YOLO versions.

Reading between the lines

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

Editorial inference: The claimed gains are measured against published AP numbers for untouched YOLO baselines; an independent comparison with baselines re-trained under the same 24-epoch recipe would clarify how much of the +0.7 to +1.0 comes from the head rather than the training setup.
Editorial inference: Neighbor Attention relies on IoU overlap, so an isolated object with no overlapping proposals receives no local-context signal; testing semantic or graph-based neighbor definitions would show whether the mechanism generalizes.
Editorial inference: The same gated decoupling of global summary, proximity context, and deformable detail could transfer to other dense prediction tasks such as instance segmentation or keypoint detection, but the paper does not test these settings.

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.

Desk Editor's Note

read the letter

You should know this paper for one reason: it proposes a genuinely different way to assemble a detection head, and if the claim holds, it is practically useful. The head combines three attention patterns—block (global), neighbor (IoU-restricted local), and deformable (detail)—fused by a lightweight gate. That combination is new, even though each ingredient is from prior work. The paper is also clearly written; the math is simple and the ablations give some support for the core intuition that unrestricted self-attention in the head is wasteful. The observation that IoU-restricted neighbor attention beats full self-attention by a small margin is the most interesting result in the paper.

The soft spots are real and they sit exactly on the headline number. Table 3 compares CSDN after 24 epochs of fine-tuning against published YOLO checkpoints that were not fine-tuned at all. Any part of the +0.7–1.0 AP could come from the extra training, not from the architecture. There are no seeds, no error bars, no same-schedule baselines. Layer count is chosen from the test table (Table 4), and then that same configuration is reported as the final result in Table 5, so the reported number carries model-selection optimism. Given that the gains are about the size of run-to-run variance for these detectors, the central claim is not yet established. This is a correct-and-request, not a rejection: the architecture is plausible and the experimental gap is fixable.

Minor issues: the deformable-attention description cites the original Deformable ConvNets paper while actually using Deformable DETR's mechanism; and the training recipe mixes Hungarian matching with NMS, which is unusual and should be explained. The citation pattern is otherwise fine.

Who is this for? A practitioner who wants a plug-and-play head and is willing to test it themselves will get a clear recipe. A reviewer should send it back for a controlled comparison: retrain the baselines under the identical 24-epoch schedule, report at least three seeds, and use a validation split for choosing layer count. With that evidence, the paper would be worth citing; without it, the headline claim is unproven.

Recommendation: yes, send to peer review, but require the controlled experiments before acceptance.

Referee Report

5 major / 5 minor

Summary. The manuscript proposes CSDN, a plug-and-play transformer-based detection head that replaces the native heads of CNN-based detectors such as YOLO. The head decomposes attention into three modules—block attention for global context, neighbor attention for local context, and deformable attention for fine details—and fuses their outputs with a gated network. The paper claims that after only 24 epochs of fine-tuning, CSDN improves COCO AP by 0.7–1.0 points over YOLOv5x, YOLOv8x, and YOLOv11x while maintaining real-time latency around 23 ms, and that the resulting model is competitive with state-of-the-art real-time detectors. Experiments in Section 4 include ablations on attention patterns and layer depth, and comparisons in Tables 3–5.

Significance. The architectural idea is timely and potentially useful: replacing exhaustive global self-attention with sparse, semantically motivated attention patterns and a learned gate could offer a practical accuracy-efficiency trade-off for real-time detectors. The plug-and-play framing is attractive, and the paper correctly identifies a known inefficiency in DETR-style heads. If the claimed gains survive controlled evaluation, this would be a meaningful contribution to the real-time detection literature. However, the current experimental validation does not yet establish the headline claim: baselines are not retrained under the same protocol, there are no error bars, and the final configuration is selected from test-set numbers. The contribution is therefore best judged as promising but not yet demonstrated at journal-level rigor.

major comments (5)

[§3.4 and §4.2, Table 3] The baseline numbers in Table 3 are taken from published checkpoints rather than from models retrained or fine-tuned under the same protocol as the CSDN rows. Since each '+CSDN' row receives 24 additional epochs of fine-tuning, the reported gains are confounded by extra training; a controlled comparison (fine-tuning the original YOLO heads for the same 24 epochs with the same optimizer, augmentation, and label assignment) is required before attributing the +0.7–1.0 AP to the CSDN architecture.
[§4.2, Tables 3–4] Every reported AP is a single run, with no seeds, standard deviations, or a description of the validation/test split. In COCO object detection, run-to-run variation for the same architecture is often on the order of 0.2–0.5 AP, which is comparable to the claimed gains. The authors should report mean and standard deviation over at least three seeds for both the CSDN and control configurations.
[§4.2–4.3, Tables 4–5] The network configuration (2/4/6 layers) is selected from COCO AP values, i.e., from the test set, and the text explicitly states 'we use 6 layers for SOTA comparisons to maximize competitiveness.' This post-hoc selection on the test set inflates the reported AP and makes the headline comparison optimistic. Layer choice should be made on a held-out validation split, and the final test result reported only for the pre-selected configuration; Table 5's use of 6 layers while Table 4 shows 4 layers at equal AP and lower latency also needs clarification.
[§3, Figure 4] The gating fusion mechanism is not specified formally. The paper states that outputs of block, neighbor, and deformable attention are combined by a 'gated neural network,' but gives no equation, input/output dimensions, normalization, or training details for the gates. Since gated aggregation is a central contribution and is ablated in Table 2, a precise specification is necessary for reproducibility.
[§3.4] The training protocol is internally inconsistent regarding label assignment and NMS: the text says training follows RT-DETR using Hungarian matching, but RT-DETR-style one-to-one assignment does not require NMS, while the paper still applies NMS with thresholds 0.25/0.6. The authors need to specify the exact label assignment (one-to-many from YOLO, or one-to-one), the loss weights, and the inference post-processing used for each row, since these choices materially affect the comparisons.

minor comments (5)

[Abstract and §4] The abstract and introduction claim the head can replace heads of 'YOLO and ResNet-based models,' but experiments cover only YOLO backbones; the claimed generality for ResNet-based detectors is not demonstrated.
[§3.2, Eq. (2)] Equation (2) defines neighbor attention via IoU(bi, bj) > 0, but the paper does not specify how boxes bi are obtained for queries at the first attention layer or for queries that have no current box prediction; if boxes come from the previous decoder layer, this should be stated.
[Table 5] Table 5 reports latencies without specifying the GPU and batch size; given the paper's real-time claims, the hardware and inference settings must be stated for each row.
[References] The reference list contains duplicate/inconsistent entries for YOLOv12 ([21] and [22] list different author sets for the same arXiv ID and title), and YOLOv11's entry ([7]) should include the version; these citation issues should be cleaned.
[Table 2] In Table 2, n+d has identical mAP50 to n-d (71.1), so the gating fusion by itself does not improve over stacking for this pair; the text should be phrased accordingly to avoid overstating the gating benefit.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are empirical measurements on COCO, not predictions derived from fitted constants or from load-bearing self-citations.

full rationale

The paper does not contain a formal derivation chain whose output is its own input. The central claims—CSDN improves AP over YOLO baselines—are supported by reported COCO mAP values obtained by training and evaluating the proposed head. These are measurements against held-out data, not predictions computed from fitted parameters, so none of the seven circularity patterns applies. There are no load-bearing self-citations: the sole author cites external prior work (DETR, Deformable DETR, RT-DETR, NSA, etc.) for components, but the paper's own gating design is not justified by an appeal to a self-cited uniqueness theorem or ansatz. One passage deserves explicit flagging: Section 4.2 states 'we use 6 layers for SOTA comparisons to maximize competitiveness,' right after Table 4 shows 4 layers matching AP 55.6 at lower latency; this is an admitted selection of the final configuration based on the benchmark metric, which is a model-selection and optimism concern, not circularity, because the reported AP is still an independent evaluation of the selected architecture. Likewise, Table 3 compares CSDN after 24-epoch fine-tuning against published baseline checkpoints, which is a comparability and statistical-validity risk, not a case of the output being equal to the input by construction. The appropriate finding is therefore no significant circularity, score 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical or mathematical entities. It relies on learned network parameters plus a small set of hand-chosen hyperparameters and domain assumptions about the COCO benchmark, IoU adjacency, and comparability of published baselines. The main load-bearing choices are the untested baseline comparability and the test-driven selection of the final configuration.

free parameters (3)

Number of CSDN layers in final model = 6 (4 layers give equal AP)
Chosen by hand from {2,4,6} based on COCO test results; used for SOTA comparisons to maximize competitiveness (Section 4.2), directly affecting reported AP and latency.
Fine-tuning length = 24 epochs
Fixed without ablated comparison; the claimed minimal fine-tuning depends on this choice (Section 3.4).
NMS thresholds = confidence 0.25, IoU 0.6
Hand-set constants used for all final predictions; their values affect reported precision and recall (Section 3.4).

assumptions (5)

standard math Softmax attention with learned projections is an appropriate feature aggregation operator.
Equations (1) and (2) build on the standard attention operator from the Transformer literature without modification.
domain assumption Published YOLO baseline AP values are comparable to the paper's own 24-epoch fine-tuning runs.
The claimed +0.7 to +1.0 AP gains in Tables 3 and 5 assume external numbers can stand in for same-protocol baselines.
domain assumption IoU-overlapping bounding boxes define semantically useful neighbor sets for local context.
Neighbor Attention masks all non-overlapping queries as irrelevant (Section 3.2, Eq. 2); no evidence is given that IoU adjacency captures the context the paper attributes to it.
domain assumption The top-level FPN feature map F_top provides sufficient global context for every ROI.
Block Attention cross-attends only to F_top (Section 3.1, Eq. 1), assuming one coarse global map can summarize scene context for all queries.
domain assumption COCO val and test metrics are an adequate measure of detection performance for the claimed real-time improvements.
All experiments use COCO following standard protocols; the real-time claim relies on reported Titan V latencies rather than a formal latency benchmark.

how reviews work

0 comments

Cite this review

Pith. "Pith review of CSDN: A Context-Gated Self-Adaptive Detection Network for Real-Time Object Detection." pith.science (2026). https://pith.science/paper/BT2NB2CQ

@misc{pith2026250617679,
  author       = {Pith},
  title        = {Pith review of: CSDN: A Context-Gated Self-Adaptive Detection Network for Real-Time Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BT2NB2CQ}},
  note         = {Machine review of arXiv:2506.17679}
}

read the original abstract

Convolutional neural networks (CNNs) have long been the cornerstone of target detection, but they are often limited by limited receptive fields, which hinders their ability to capture global contextual information. We re-examined the DETR-inspired detection head and found substantial redundancy in its self-attention module. To solve these problems, we introduced the Context-Gated Scale-Adaptive Detection Network (CSDN), a Transformer-based detection header inspired by human visual perception: when observing an object, we always concentrate on one site, perceive the surrounding environment, and glance around the object. This mechanism enables each region of interest (ROI) to adaptively select and combine feature dimensions and scale information from different patterns. CSDN provides more powerful global context modeling capabilities and can better adapt to objects of different sizes and structures. Our proposed detection head can directly replace the native heads of various CNN-based detectors, and only a few rounds of fine-tuning on the pre-trained weights can significantly improve the detection accuracy.

Figures

Figures reproduced from arXiv: 2506.17679 by the authors.

**Figure 2.** Illustration simulating human visual focus. Identifying [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** 2D extensions of NSA attention patterns, Yellow areas [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: The complete architecture of CSDN, utilizing Gated Fu [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

**Figure 6.** Figure 6: Neighbor Attention mechanism. The central query’s [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]

**Figure 5.** Figure 5: Block Attention mechanism. The input image’s multi [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

**Figure 7.** Figure 7: Deformable Attention mechanism. Learned offsets ad [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

28 extracted references · 19 canonical work pages

[1]

Native sparse attention: Hardware- aligned and natively trainable sparse attention

DeepSeek AI. Native sparse attention: Hardware- aligned and natively trainable sparse attention. ArXiv, abs/2502.11089, 2025. [1] in text. 3, 6

arXiv 2025
[2]

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. A prominent YOLO variant. 1

arXiv 2004
[3]

YOLOv11x + Ours vs

Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Table 5. YOLOv11x + Ours vs. Other Real-time Object Detectors on COCO (using the largest scale for each model without pre-trained on Object360). AP refers to mAP@[.5:.95]. Category Method AP AP50 Params(M) GFLOPs Latency(ms) YOLO Series YOLOv10x [24] 54.5 71.0 31.7 170.6 14.8 YOLOv11x (Baseline) [...

work page 2020
[4]

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. In Computer Vision – ECCV 2020, pages 213–229, 2020. Cited as [10] in Intro- duction and CITE DETR SERIES in Introduction and [4] in text. 3, 7

work page 2020
[5]

Deformable convolutional networks

Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. In International Conference on Computer Vision (ICCV), 2017. Original deformable convolutions. 2

work page 2017
[6]

Relation networks for object detection

Han Hu, Jiayuan Gu, Zheng Zhang, Jifeng Dai, and Yichen Wei. Relation networks for object detection. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. Cited as [2] in Methodology. 4

work page 2018
[7]

Ultralytics yolo11

Glenn Jocher and Jing Qiu. Ultralytics yolo11. GitHub Repository, https://github.com/ultralytics/ ultralytics, 2024. 7

work page 2024
[8]

Lawrence Zitnick

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C. Lawrence Zitnick. Microsoft COCO: Common objects in context. In European Conference on Computer Vision (ECCV) , 2014. The COCO dataset paper. 4, 5

work page 2014

Show all 28 references

[9]

Feature pyramid networks for object detection

Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017. Cited as [1] in Methodology. 1

2017
[10]

Girshick, Kaiming He, and Piotr Doll ´ar

Tsung-Yi Lin, Priya Goyal, Ross B. Girshick, Kaiming He, and Piotr Doll ´ar. Focal loss for dense object detection. In 2017 IEEE International Conference on Computer Vision (ICCV), pages 2999–3007, 2017. Cited as [8] in Training Method and [10] in text. 5, 7

2017
[11]

Decoupled weight decay regularization

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. ArXiv, abs/1711.05101, 2017. Cited as [10] in Training Method and [11] in text. Corrected year to 2017 for original arXiv submission, though ICLR publication was

2017 arXiv
[12]

Rt-detr: Detr-like transformers for end-to-end ob- ject detection

Wenyu Lv, Shangliang Xu, Yian Zhao, Guanzhong Wang, Jinman Wei, Cheng Cui, Yuning Du, Qingqing Dang, and Yi Liu. Rt-detr: Detr-like transformers for end-to-end ob- ject detection. ArXiv, abs/2304.08041, 2023. Cited as [6] in Training Method and [12] in text. 3, 5, 6, 7

2023 arXiv
[13]

Rtmdet: An empirical study of designing real-time object detectors

Chengqi Lyu, Wenwei Zhang, Haian Huang, Yue Zhou, Yudong Wang, Yanyi Liu, Shilong Zhang, and Kai Chen. Rtmdet: An empirical study of designing real-time object detectors. arXiv preprint arXiv:2212.07784, 2022. 6, 7

2022 arXiv
[14]

Zabairu, Yan P

Haodong Ouyang, Kun Wang, Guodong Wang, Shuaibu S. Zabairu, Yan P. Lin, Hongkai Yu, and Feng Lin. Deyo: Detr with yolo for end-to-end object detection. ArXiv, abs/2402.16370, 2024. Cited as [5] in Training Method and

2024 arXiv
[15]

D-fine: Redefine regression task of de- trs as fine-grained distribution refinement

Yansong Peng, Hebei Li, Peixi Wu, Yueyi Zhang, Xiaoyan Sun, and Feng Wu. D-fine: Redefine regression task of de- trs as fine-grained distribution refinement. arXiv preprint arXiv:2410.13842, 2024. 7

2024 arXiv
[16]

Yolo9000: Better, faster, stronger

Joseph Redmon and Ali Farhadi. Yolo9000: Better, faster, stronger. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 6517–6525, 2017. [16] in text. 2, 7

2017
[17]

Yolov3: An incremental improvement

Joseph Redmon and Ali Farhadi. Yolov3: An incremental improvement. ArXiv, abs/1804.02767, 2018. [17] in text. 2, 7

2018 arXiv
[18]

You only look once: Unified, real-time object detec- tion

Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detec- tion. In Conference on Computer Vision and Pattern Recog- nition (CVPR), 2016. The original YOLO paper. 1

2016
[19]

Girshick, and Ali Farhadi

Joseph Redmon, Santosh Kumar Divvala, Ross B. Girshick, and Ali Farhadi. You only look once: Unified, real-time ob- ject detection. In 2016 IEEE Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 779–788, 2016. Cited as [6, 7] in Introduction, [13] in Introd...

2016
[20]

Reid, and Silvio Savarese

Hamid Rezatofighi, Nathan Tsoi, JunYoung Gwak, Amir Sadeghian, Ian D. Reid, and Silvio Savarese. Generalized in- tersection over union: A metric and a loss for bounding box regression. In 2019 IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), pages 658–6...

2019
[22]

Yolov12: Attention-centric real-time object detectors

Yunjie Tian, Qixiang Ye, and David Doermann. Yolov12: Attention-centric real-time object detectors. arXiv preprint arXiv:2502.12524, 2025. 2, 7

2025 arXiv
[23]

Gomez, Ł ukasz Kaiser, and Illia Polosukhin

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N. Gomez, Ł ukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neu- ral Information Processing Systems (NeurIPS) , 2017. The original Transformer paper. 1

2017
[24]

Yolov10: Real-time end- to-end object detection

Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jun- gong Han, and Guiguang Ding. Yolov10: Real-time end- to-end object detection. arXiv preprint arXiv:2405.14458,

arXiv
[25]

Gold-yolo: Ef- ficient object detector via gather-and-distribute mechanism

Chengcheng Wang, Wei He, Ying Nie, Jianyuan Guo, Chuanjian Liu, Kai Han, and Yunhe Wang. Gold-yolo: Ef- ficient object detector via gather-and-distribute mechanism. arXiv preprint arXiv:2309.11331, 2023. 7

2023 arXiv
[26]

Ni, and Heung-Yeung Shum

Hao Zhang, Feng Li, Shilong Liu, Lei Zhang, Hang Su, Jun Zhu, Lionel M. Ni, and Heung-Yeung Shum. Dino: Detr with improved denoising anchor boxes for end-to-end object detection. 2023. 3

2023
[27]

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. CVPR 2024,

2024
[28]

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. In International Conference on Learning Representations (ICLR) , 2021. Cited as [3] in Deformable Attention and [11] in Introductio...

2021
[2025]

[21] in text. 2, 6, 8

Pith tools

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

[1] [1]

Native sparse attention: Hardware- aligned and natively trainable sparse attention

DeepSeek AI. Native sparse attention: Hardware- aligned and natively trainable sparse attention. ArXiv, abs/2502.11089, 2025. [1] in text. 3, 6

arXiv 2025

[2] [2]

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. A prominent YOLO variant. 1

arXiv 2004

[3] [3]

YOLOv11x + Ours vs

Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Table 5. YOLOv11x + Ours vs. Other Real-time Object Detectors on COCO (using the largest scale for each model without pre-trained on Object360). AP refers to mAP@[.5:.95]. Category Method AP AP50 Params(M) GFLOPs Latency(ms) YOLO Series YOLOv10x [24] 54.5 71.0 31.7 170.6 14.8 YOLOv11x (Baseline) [...

work page 2020

[4] [4]

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. In Computer Vision – ECCV 2020, pages 213–229, 2020. Cited as [10] in Intro- duction and CITE DETR SERIES in Introduction and [4] in text. 3, 7

work page 2020

[5] [5]

Deformable convolutional networks

Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. In International Conference on Computer Vision (ICCV), 2017. Original deformable convolutions. 2

work page 2017

[6] [6]

Relation networks for object detection

Han Hu, Jiayuan Gu, Zheng Zhang, Jifeng Dai, and Yichen Wei. Relation networks for object detection. In Proceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018. Cited as [2] in Methodology. 4

work page 2018

[7] [7]

Ultralytics yolo11

Glenn Jocher and Jing Qiu. Ultralytics yolo11. GitHub Repository, https://github.com/ultralytics/ ultralytics, 2024. 7

work page 2024

[8] [8]

Lawrence Zitnick

Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C. Lawrence Zitnick. Microsoft COCO: Common objects in context. In European Conference on Computer Vision (ECCV) , 2014. The COCO dataset paper. 4, 5

work page 2014