Pith. sign in

REVIEW 3 major objections 6 minor 62 references

DS-Det: Single-Query Paradigm and Attention Disentangled Learning for Flexible Object Detection

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read DS-Det argues that separating the decoder's cross-attention (locating) from self-attention (deduplication), together with image-dependent query selection, lets a transformer detector beat the fixed-query DINO baseline by +1.5% AP on COCO…

desk verdict A solid, carefully ablated DETR variant with a genuinely flexible query mechanism; the central claims hold, though the fixed threshold S deserves one more cross-domain sensitivity test. read the letter →

arxiv 2507.19807 v1 pith:UUV3II6H submitted 2025-07-26 cs.CV

Guiping Cao , Xiangyuan Lan , Wenjian Huang , Jianguo Zhang , Dongmei Jiang , Yaowei Wang This is my paper · ORCID
classification cs.CV
keywords flexibleobjectdetectionsingle-queryparadigmattentiondisentangledlearningqueryambiguityDETRdecodersmallefficienttransformer
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 claims that standard DETR-style detectors hurt themselves by interleaving two opposing attention operations layer by layer: self-attention pushes queries apart while cross-attention pulls them together around the same object, and mixing one-to-one with one-to-many label assignments in shared decoder layers makes each query unsure whether to predict a single object or one of many. To remove this 'query ambiguity', DS-Det lets the encoder choose a flexible number of queries per image and splits decoding into a box-locating stage that uses only cross-attention and a deduplication stage that uses self-attention. If correct, a simpler decoder can be both more accurate and more efficient than existing fixed-query designs, with gains concentrated on small objects and with the query count adapting to the image. The paper demonstrates this on COCO, WiderPerson, and CrowdHuman across five backbones and also shows the recipe transfers to Deformable DETR.

What carries the argument

The load-bearing object is the Attention Disentangled Decoder (ADD), built on the new Single-Query paradigm. FLET derives the queries by applying a classification head to all encoder tokens and selecting those above a threshold S, so the number of queries is flexible and tied to image content rather than a fixed hyperparameter. ADD then separates the attention roles: the Box Locating Part contains only cross-attention and uses one-to-many matching (each ground-truth box replicated K times) to locate objects, while the Deduplication Part contains multi-head self-attention blocks and uses one-to-one matching to suppress duplicate predictions, with a stop-gradient preventing the two matching schemes from conflicting. A third component, the PoCoo loss, reweights the classification loss by box size so small objects receive higher learning priority.

What would settle it

Measure the calibration of the FLET classification head on a held-out domain (for example, aerial imagery with hundreds of small objects per image) and compare AP at S=0.02 against an oracle that selects the optimal query count per image; if the fixed threshold consistently picks too few or too many queries, the flexibility claim reverses.

Watch

Extended reading notes

Core claim

The central discovery is that inefficient training and weak decoders in DETR-like detectors trace back to two entangled sources: the recurrent opposing interactions between cross-attention and self-attention, and the mixing of one-to-one and one-to-many matching within shared-weight decoder layers. DS-Det resolves both by introducing a Flexible Single-Query generation (FLET) module, which runs a classification head over all encoder tokens and uses a threshold to form an image-dependent pool of queries, and by redesigning the decoder as an Attention Disentangled Decoder (ADD). In ADD, the Box Locating Part uses only cross-attention with one-to-many matching to locate objects, while the Deduplication Part uses self-attention with one-to-one matching to suppress duplicates, with a stop-gradient between the two parts so their supervision does not conflict. On COCO, DS-Det with ResNet50 outperforms the DINO baseline by +1.5% AP and +2.3% AP_S under a 12-epoch schedule, improves decoder inference speed by 34.8% at 900 queries, and scales its query count with the number of objects in the image (about 65 queries on sparse images, up to 900 on dense ones).

Load-bearing premise

The load-bearing premise is that the FLET classification head's scores reliably separate object-bearing encoder tokens from background, so a fixed threshold S=0.02 chooses roughly the right number of queries for any image.

Editorial extensions

If this is right

  • A decoder with only 4 cross-attention-only layers plus 2 deduplication layers can outperform a 6-layer mixed-attention decoder on the same 12-epoch training budget.
  • Sparse images need far fewer queries: the model uses roughly 65 queries on images with 1-5 objects versus the fixed 900 in DINO, with comparable or better AP.
  • Decoder inference is faster (9.2 ms vs 14.1 ms at 900 queries), because self-attention layers are concentrated in only two of the six decoder stages.
  • Small-object AP improves (+2.3% AP_S over DINO at 12 epochs), which the paper attributes to the box-size-aware PoCoo loss.
  • The recipe is portable: applying FLET, ADD, and PoCoo to Deformable DETR raises its AP by +3.1% at 30 epochs versus the original 50-epoch result.

Reading between the lines

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

  • The paper's diagnosis of 'query ambiguity' suggests a general design rule for set-prediction decoders: whenever one-to-many and one-to-one supervision share weights, separate them by stage or by gradient, not just by loss weighting.
  • Because the threshold S controls the query budget at inference without retraining, DS-Det could be deployed with an adjustable speed-accuracy trade-off, for example raising S on resource-constrained devices or lowering it for dense scenes.
  • The flexible query count is especially promising for open-ended or language-coupled detection, where feeding 65 rather than 900 queries to a language model would cut downstream computation by roughly an order of magnitude.
  • A testable extension is applying FLET and ADD to panoptic or video detection, where the number of objects per frame is even more variable and fixed-query decoders waste computation on empty frames.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes DS-Det, a DETR-family detector that replaces fixed content/positional queries with a flexible set of queries produced by a FLET module (encoder-token classification followed by threshold selection S), partitions the decoder into a Box Locating Part (cross-attention with one-to-many matching) and a Deduplication Part (self-attention with one-to-one matching), and uses a PoCoo loss that reweights classification by box size. Experiments on COCO, WiderPerson, and CrowdHuman across five backbones report consistent AP gains over DINO (e.g., +1.5 AP at 12 epochs on COCO with ResNet50), a 34.8% decoder speedup at 900 queries, and an additional Deformable-DETR variant in the appendix.

Significance. If the empirical results hold, the paper makes a useful contribution: it shows that a decoder with fewer self-attention layers and a single query type can match or beat a fixed-query mixed-attention decoder on equal training budgets, and it provides extensive ablations (component, loss, SA count, threshold, layer split, matching order, cost weights), speed measurements, and cross-backbone/cross-dataset experiments. The FLOPs and query-count analysis, the COCO object-number subsets, and the downstream open-ended detection complexity test are concrete evidence for the flexibility claim. The main risks are the threshold-based query selection's transferability and a train/inference mismatch in query construction, both of which affect the central 'flexible query' claim.

major comments (3)
  1. [Sec. 3.3 and Appendix B.4 (Table 14)] The flexible-query mechanism is controlled by the FLET threshold S=0.02, but the only sensitivity analysis is on COCO val, where AP moves only from 50.2 to 50.6 for S in [0.01,0.05]. The same fixed S is used without additional analysis on WiderPerson (Table 4) and CrowdHuman (Table 20), and no query-count statistics or calibration evidence are reported for these datasets. Because the paper's accuracy/efficiency claim is specifically that FLET selects the right number of queries per image, a score-distribution shift in a new domain could select too many or too few queries and reverse the reported gains; the WiderPerson margins (e.g., +0.49 AP for ResNet50 in Table 4) are small enough that this is a realistic risk. Please provide per-dataset AP-vs-S sweeps (or an alternative adaptive selection criterion) and report the distribution of selected query counts on WiderPerson and CrowdHuman.
  2. [Sec. 3.3 (Query Sampling and Query Alignment)] The paper does not specify the training objective of the FLET classification head or how the sampled queries used during training relate to the thresholded queries used at inference. Table 5 attributes +4.6 AP to FLET over the flexible-c baseline, so this mechanism is load-bearing; however, the reader cannot tell whether the improvement comes from threshold selection, query-sampling regularization, or placeholder-query padding. Please state the classification loss and label assignment for encoder tokens, and ablate query sampling against direct top-k/threshold selection during training.
  3. [Appendix B.4 (CrowdHuman, Table 20)] The text states that DS-Det obtains overall higher performance than DINO variants on CrowdHuman, but Table 20 shows DS-Det-Strip-MLP-T at 87.92 AP versus DINO-Strip-MLP-T at 88.38 AP, a decrease of 0.46 AP. Since the paper's general-effectiveness claim rests on consistent improvements, this exception must be acknowledged and explained (or the claim qualified) rather than glossed over.
minor comments (6)
  1. [Sec. 3.2, Eq. (2)] The second summation in Eq. (2) uses j=i-T1 with the condition i>T1, reusing i from the first summation; this indexing is confusing and should be rewritten with separate layer indices for the BLP and DP blocks.
  2. [Fig. 2] The figure caption contains the placeholder text '替换图片：', which appears to be an editing artifact and should be removed before publication.
  3. [Sec. 3.5, Eq. (6)] The PoCoo loss uses t_i in the BCE term without defining it; the paper should specify how the IA-BCE label from Align-DETR is constructed, since the formula is otherwise incomplete.
  4. [Table 3 vs. Appendix C.4] The FLOPs for DS-Det-ResNet50 are reported as 275G in Table 3 but as 273G in Appendix C.4 for the same configuration; the numbers should be reconciled.
  5. [Table 5] The row label 'flexible-c' in Table 5 is not defined in the table caption; please clarify whether it means a fixed number of queries chosen manually for flexible-conditioned training.
  6. [Sec. 3.1 and Abstract] The 'ROT' and 'query ambiguity' constructs are asserted rather than directly measured; the ablation results are consistent with the proposed story, but a direct diagnostic (e.g., query-similarity or gradient-conflict analysis) would strengthen the causal claim that the decoder design addresses these specific phenomena.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; DS-Det's claims are empirical benchmark results with independently evaluated components.

full rationale

The paper's central claims are experimental: DS-Det is reported to outperform DINO on COCO, WiderPerson, and CrowdHuman across several backbones. The FLET query-selection mechanism uses a classification head and a fixed threshold S=0.02, which is a tuned hyperparameter rather than a fitted quantity disguised as a prediction; Table 14 reports AP sensitivity over S, and the paper does not claim to derive the AP gains from the threshold. The ADD decoder separates cross-attention localization from self-attention deduplication, and its effectiveness is supported by ablations (Tables 5, 6, 9, 15, 16) rather than by an argument that reduces to its own assumptions. The PoCoo loss explicitly reweights classification by box size, but the reported small-object AP improvement is a measured outcome, not an equivalence forced by the loss definition. Self-citations such as Strip-MLP [5] and MLP-DINO [3] appear as backbone choices or related-work references, but they are not load-bearing: the reported gains are computed against independently published baselines and do not depend on an unverified self-cited theorem or uniqueness claim. The remaining concern about threshold calibration across domains is a generalization or correctness risk, not a circularity, because the paper never presents the flexible query count as a derived prediction from an input that already contains it.

Assumptions & free parameters 7 free parameters · 5 assumptions · 2 invented entities

The central empirical result relies on a standard DINO training pipeline plus several tuned hyperparameters. The main introduced constructs, ROT and query ambiguity, are explanatory labels without direct measurement, and the architecture's success is benchmarked externally rather than derived from first principles.

free parameters (7)
  • FLET threshold S = 0.02
    Controls how many encoder tokens become queries; chosen after ablation on COCO val (Table 14).
  • PoCoo exponent alpha = not reported (stated range 0-1)
    Controls box-size downweighting in the loss; the paper never states the chosen value, which hurts reproducibility.
  • one-to-many replication K = 6
    Number of copies of each ground-truth box in BLP; chosen by ablation (Table 17).
  • BLP/DP layer split = 4 BLP + 2 DP
    Number of decoder layers assigned to localization vs deduplication; chosen by ablation (Table 15).
  • SA blocks in DP (lambda) = 2
    Self-attention block count in DPMSAB; chosen by ablation (Table 6).
  • loss weights = PoCoo_enc=1.5, PoCoo_BLP=2.0, L1_BLP=5.0, GIoU_BLP=2.0, PoCoo_DP=2.0, L1_DP=5.0, GIoU_DP=2.0
    Selected via ablation (Table 19); the dependency of the reported gains on these exact values is not separately analyzed.
  • classification cost weight in BLP matching = 0.2
    Ablated in Table 18; affects which queries are matched to ground-truth boxes.
assumptions (5)
  • domain assumption The DINO training recipe and COCO/WiderPerson evaluation metrics are valid for comparing detectors.
    Used as the experimental protocol; the paper inherits all assumptions of the DINO codebase and the standard COCO metric.
  • domain assumption Encoder token classification scores correlate with object presence and informativeness.
    FLET relies on this to select queries (Sec. 3.3); no calibration analysis is given.
  • domain assumption Cross-attention gathers queries around an object while self-attention disperses them.
    Taken from DAC-DETR and used to justify the ADD design (Sec. 3.1).
  • domain assumption Deformable attention's point sampling provides enough positional information to make explicit positional queries unnecessary.
    Used to justify dropping PQ (Sec. 3.3, Table 8).
  • domain assumption One-to-one matching in self-attention layers can suppress duplicate predictions without post-processing NMS.
    Core of the DP design; supported by ablations but not by a formal argument.
invented entities (2)
  • Recurrent Opposing inTeractions (ROT)
    purpose: Postulated mechanism by which alternating SA and CA layers impede query convergence and degrade decoder efficiency.
    The paper names and repeatedly invokes ROT (Secs. 1, 3.1) but presents no direct measurement of this interaction; its support is the performance of the proposed architecture.
  • Query ambiguity
    purpose: Postulated failure mode from mixing one-to-one and one-to-many label assignments with shared decoder weights.
    Introduced as an explanatory construct (Secs. 1, 3.4); no metric quantifies it directly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DS-Det: Single-Query Paradigm and Attention Disentangled Learning for Flexible Object Detection." pith.science (2026). https://pith.science/paper/UUV3II6H

@misc{pith2026250719807,
  author       = {Pith},
  title        = {Pith review of: DS-Det: Single-Query Paradigm and Attention Disentangled Learning for Flexible Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UUV3II6H}},
  note         = {Machine review of arXiv:2507.19807}
}
read the original abstract

Popular transformer detectors have achieved promising performance through query-based learning using attention mechanisms. However, the roles of existing decoder query types (e.g., content query and positional query) are still underexplored. These queries are generally predefined with a fixed number (fixed-query), which limits their flexibility. We find that the learning of these fixed-query is impaired by Recurrent Opposing inTeractions (ROT) between two attention operations: Self-Attention (query-to-query) and Cross-Attention (query-to-encoder), thereby degrading decoder efficiency. Furthermore, "query ambiguity" arises when shared-weight decoder layers are processed with both one-to-one and one-to-many label assignments during training, violating DETR's one-to-one matching principle. To address these challenges, we propose DS-Det, a more efficient detector capable of detecting a flexible number of objects in images. Specifically, we reformulate and introduce a new unified Single-Query paradigm for decoder modeling, transforming the fixed-query into flexible. Furthermore, we propose a simplified decoder framework through attention disentangled learning: locating boxes with Cross-Attention (one-to-many process), deduplicating predictions with Self-Attention (one-to-one process), addressing "query ambiguity" and "ROT" issues directly, and enhancing decoder efficiency. We further introduce a unified PoCoo loss that leverages box size priors to prioritize query learning on hard samples such as small objects. Extensive experiments across five different backbone models on COCO2017 and WiderPerson datasets demonstrate the general effectiveness and superiority of DS-Det. The source codes are available at https://github.com/Med-Process/DS-Det/.

Figures

Figures reproduced from arXiv: 2507.19807 by the authors.

Figure 1
Figure 1. The mixing of one-to-one and one-to-many matching with shared weights in the decoder introduces [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The overall architecture of the DS-Det model. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The main pipeline of the FLET module. cross-attention with one-to-many matching for localization and employing self-attention with one-to-one matching for recognition. Box Locating Part (BLP). The BLP is specifically designed to locate potential objects accurately through a one-to-many matching mechanism using multiple queries. This is achieved by replicating each ground truth box𝐾 times and aligning each replicated… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: The number of queries and FLOPs of the model w.r.t threshold of classification score [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: The relation between the number of objects and queries dynamically generated by the DS-Det. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The performance comparison on the subsets of val2017 of COCO. The subsets are generated based on the number of [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: The histogram of the WiderPerson validation dataset. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Convergence curves of Deformable DETR and Deformable DETR-DS model. [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Comparison of transformer-based models of AP and AP [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Visualization of queries in simple scenarios. The number of queries (Q-Num) selected for each image is on the top [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Visualization of queries in complex scenarios. [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]
Figure 12
Figure 12. Figure 12: Visualization of queries in complex scenarios. [PITH_FULL_IMAGE:figures/full_fig_p024_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 40 canonical work pages

  1. [1]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. 2016. Layer normaliza- tion. arXiv preprint arXiv:1607.06450 (2016)

    arXiv 2016

  2. [2]

    Zhi Cai, Songtao Liu, Guodong Wang, Zheng Ge, Xiangyu Zhang, and Di Huang

    work page

  3. [3]

    Guiping Cao, Wenjian Huang, Xiangyuan Lan, Jianguo Zhang, Dongmei Jiang, and Yaowei Wang. 2024. MLP-DINO: Category Modeling and Query Graphing with Deep MLP for Object Detection. In Proceedings of the Thirty-Third Interna- tional Joint Conference on Artificial Intelligence . 605–613

    work page 2024

  4. [4]

    Guiping Cao, Wenjian Huang, Xiangyuan Lan, Jianguo Zhang, Dongmei Jiang, and Yaowei Wang. 2025. Cross-DINO: Cross the Deep MLP and Transformer for Small Object Detection. arXiv preprint arXiv:2505.21868 (2025)

    work page Pith review arXiv 2025

  5. [5]

    Guiping Cao, Shengda Luo, Wenjian Huang, Xiangyuan Lan, Dongmei Jiang, Yaowei Wang, and Jianguo Zhang. 2023. Strip-MLP: Efficient Token Interaction for Vision MLP. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 1494–1504

    work page 2023

  6. [6]

    Xipeng Cao, Peng Yuan, Bailan Feng, and Kun Niu. 2022. Cf-detr: Coarse-to- fine transformers for end-to-end object detection. In Proceedings of the AAAI Conference on Artificial Intelligence. 185–193

    work page 2022

  7. [7]

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexan- der Kirillov, and Sergey Zagoruyko. 2020. End-to-end object detection with transformers. In European conference on computer vision . Springer, 213–229

    work page 2020

  8. [8]

    Qiang Chen, Xiaokang Chen, Jian Wang, Shan Zhang, Kun Yao, Haocheng Feng, Junyu Han, Errui Ding, Gang Zeng, and Jingdong Wang. 2023. Group detr: Fast detr training with group-wise one-to-many assignment. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 6633–6642

    work page 2023

Show all 62 references
  1. [9]

    Shoufa Chen, Peize Sun, Yibing Song, and Ping Luo. 2023. Diffusiondet: Diffusion model for object detection. InProceedings of the IEEE/CVF International Conference on Computer Vision. 19830–19843

    2023

  2. [10]

    Gong Cheng, Xiang Yuan, Xiwen Yao, Kebing Yan, Qinghua Zeng, Xingxing Xie, and Junwei Han. 2023. Towards large-scale small object detection: Survey and benchmarks. IEEE Transactions on Pattern Analysis and Machine Intelligence (2023)

    2023

  3. [11]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv prepri...

    2020 arXiv

  4. [12]

    Zheng Ge, Zequn Jie, Xin Huang, Rong Xu, and Osamu Yoshie. 2020. Ps-rcnn: Detecting secondary human instances in a crowd via primary object suppression. In 2020 IEEE international conference on multimedia and expo (ICME) . IEEE, 1–6

    2020

  5. [13]

    Ross Girshick. 2015. Fast r-cnn. In Proceedings of the IEEE international conference on computer vision. 1440–1448

    2015

  6. [14]

    Albert Gu and Tri Dao. 2023. Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752 (2023)

    2023 arXiv

  7. [15]

    Meng-Hao Guo, Zheng-Ning Liu, Tai-Jiang Mu, and Shi-Min Hu. 2023. Beyond Self-Attention: External Attention Using Two Linear Layers for Visual Tasks.IEEE Transactions on Pattern Analysis & Machine Intelligence 45, 05 (2023), 5436–5447

    2023

  8. [16]

    Kaiming He, Georgia Gkioxari, Piotr Dollár, and Ross Girshick. 2017. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision . 2961–2969

    2017

  9. [17]

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

    2016

  10. [18]

    Yuzhe He, Ning He, Ren Zhang, Kang Yan, and Haigang Yu. 2022. Multi-scale feature balance enhancement network for pedestrian detection. Multimedia Systems 28, 3 (2022), 1135–1145

    2022

  11. [19]

    Zhengdong Hu, Yifan Sun, Jingdong Wang, and Yi Yang. 2024. DAC-DETR: Divide the attention layers and conquer. Advances in Neural Information Processing Systems 36 (2024)

    2024

  12. [20]

    Yi-Xin Huang, Hou-I Liu, Hong-Han Shuai, and Wen-Huang Cheng. 2024. Dq- detr: Detr with dynamic query for tiny object detection. In European Conference on Computer Vision. Springer, 290–305

    2024

  13. [21]

    Ding Jia, Yuhui Yuan, Haodi He, Xiaopei Wu, Haojun Yu, Weihong Lin, Lei Sun, Chao Zhang, and Han Hu. 2023. Detrs with hybrid matching. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 19702–19712

    2023

  14. [22]

    Rahima Khanam and Muhammad Hussain. 2024. YOLOv11: An Overview of the Key Architectural Enhancements. arXiv preprint arXiv:2410.17725 (2024)

    2024 arXiv

  15. [23]

    Feng Li, Hao Zhang, Shilong Liu, Jian Guo, Lionel M Ni, and Lei Zhang. 2022. Dn-detr: Accelerate detr training by introducing query denoising. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 13619– 13627

    2022

  16. [24]

    Chuang Lin, Yi Jiang, Lizhen Qu, Zehuan Yuan, and Jianfei Cai. 2024. Generative Region-Language Pretraining for Open-Ended Object Detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 13958– 13968

    2024

  17. [25]

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. 2014. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proce...

    2014

  18. [26]

    Shilong Liu, Feng Li, Hao Zhang, Xiao Yang, Xianbiao Qi, Hang Su, Jun Zhu, and Lei Zhang. 2022. Dab-detr: Dynamic anchor boxes are better queries for detr. arXiv preprint arXiv:2201.12329 (2022)

    2022 arXiv

  19. [27]

    Shilong Liu, Tianhe Ren, Jiayu Chen, Zhaoyang Zeng, Hao Zhang, Feng Li, Hongyang Li, Jun Huang, Hang Su, Jun Zhu, et al. 2023. Detection Transformer with Stable Matching. arXiv preprint arXiv:2304.04742 (2023)

    2023 arXiv

  20. [28]

    Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, et al. 2024. Grounding dino: Marry- ing dino with grounded pre-training for open-set object detection. In European conference on computer vision . Springer, 38–55

    2024

  21. [29]

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. 2016. Ssd: Single shot multibox detec- tor. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part...

    2016

  22. [30]

    Yang Liu, Peng Sun, Nickolas Wergeles, and Yi Shang. 2021. A survey and performance evaluation of deep learning methods for small object detection. Expert Systems with Applications 172 (2021), 114602

    2021

  23. [31]

    Yue Liu, Yunjie Tian, Yuzhong Zhao, Hongtian Yu, Lingxi Xie, Yaowei Wang, Qixiang Ye, Jianbin Jiao, and Yunfan Liu. 2024. Vmamba: Visual state space model. Advances in neural information processing systems 37 (2024), 103031–103063

    2024

  24. [32]

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. 2021. Swin transformer: Hierarchical vision transformer us- ing shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision. 10012–10022

    2021

  25. [33]

    Chunjie Ma, Li Zhuo, Jiafeng Li, Yutong Zhang, and Jing Zhang. 2023. Cascade Transformer Decoder Based Occluded Pedestrian Detection With Dynamic De- formable Convolution and Gaussian Projection Channel Attention Mechanism. IEEE Transactions on Multimedia (2023)

    2023

  26. [34]

    Depu Meng, Xiaokang Chen, Zejia Fan, Gang Zeng, Houqiang Li, Yuhui Yuan, Lei Sun, and Jingdong Wang. 2021. Conditional detr for fast training convergence. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 3651– 3660

    2021

  27. [35]

    Jeffrey Ouyang-Zhang, Jang Hyun Cho, Xingyi Zhou, and Philipp Krähenbühl

  28. [36]

    Yifan Pu, Weicong Liang, Yiduo Hao, Yuhui Yuan, Yukang Yang, Chao Zhang, Han Hu, and Gao Huang. 2024. Rank-DETR for high quality object detection. Advances in Neural Information Processing Systems 36 (2024)

    2024

  29. [37]

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. 2016. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition . 779–788

    2016

  30. [38]

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. 2015. Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems 28 (2015)

    2015

  31. [39]

    Tianhe Ren, Shilong Liu, Feng Li, Hao Zhang, Ailing Zeng, Jie Yang, Xingyu Liao, Ding Jia, Hongyang Li, He Cao, et al . 2023. detrex: Benchmarking Detection Transformers. arXiv preprint arXiv:2306.07265 (2023)

    2023 arXiv

  32. [40]

    Byungseok Roh, JaeWoong Shin, Wuhyun Shin, and Saehoon Kim. 2021. Sparse detr: Efficient end-to-end object detection with learnable sparsity. arXiv preprint arXiv:2111.14330 (2021)

    2021 arXiv

  33. [41]

    Danila Rukhovich, Konstantin Sofiiuk, Danil Galeev, Olga Barinova, and An- ton Konushin. 2021. Iterdet: iterative scheme for object detection in crowded environments. In Structural, Syntactic, and Statistical Pattern Recognition: Joint IAPR International Workshops, S+ SSPR 202...

    2021

  34. [42]

    Shuai Shao, Zijian Zhao, Boxun Li, Tete Xiao, Gang Yu, Xiangyu Zhang, and Jian Sun. 2018. Crowdhuman: A benchmark for detecting human in a crowd. arXiv preprint arXiv:1805.00123 (2018)

    2018 arXiv

  35. [43]

    Zhuoran Shen, Mingyuan Zhang, Haiyu Zhao, Shuai Yi, and Hongsheng Li

  36. [44]

    Yao Teng, Haisong Liu, Sheng Guo, and Limin Wang. 2023. Stageinteractor: Query-based object detector with cross-stage interaction. In Proceedings of the MM ’25, October 27–31, 2025, Dublin, Ireland Cao et al. IEEE/CVF International Conference on Computer Vision . 6577–6588

    2023

  37. [45]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017)

    2017

  38. [46]

    Shuai Wang, Yao Teng, and Limin Wang. 2023. Deep equilibrium object detection. In Proceedings of the IEEE/CVF international conference on computer vision . 6296– 6306

    2023

  39. [47]

    Yingming Wang, Xiangyu Zhang, Tong Yang, and Jian Sun. 2022. Anchor detr: Query design for transformer-based detector. In Proceedings of the AAAI confer- ence on artificial intelligence . 2567–2575

    2022

  40. [48]

    Zhuyu Yao, Jiangbo Ai, Boxun Li, and Chi Zhang. 2021. Efficient detr: improving end-to-end object detector with dense prior. arXiv preprint arXiv:2104.01318 (2021)

    2021 arXiv

  41. [49]

    Weihao Yu and Xinchao Wang. 2025. Mambaout: Do we really need mamba for vision?. In Proceedings of the Computer Vision and Pattern Recognition Conference . 4484–4496

    2025

  42. [50]

    Gongjie Zhang, Zhipeng Luo, Yingchen Yu, Kaiwen Cui, and Shijian Lu. 2022. Accelerating DETR convergence via semantic-aligned matching. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 949–958

    2022

  43. [51]

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

    2022 arXiv

  44. [52]

    Shilong Zhang, Xinjiang Wang, Jiaqi Wang, Jiangmiao Pang, Chengqi Lyu, Wen- wei Zhang, Ping Luo, and Kai Chen. 2023. Dense distinct query for end-to-end object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7329–7338

    2023

  45. [53]

    Shifeng Zhang, Yiliang Xie, Jun Wan, Hansheng Xia, Stan Z Li, and Guodong Guo. 2019. Widerperson: A diverse dataset for dense pedestrian detection in the wild. IEEE Transactions on Multimedia 22, 2 (2019), 380–393

    2019

  46. [54]

    Chuyang Zhao, Yifan Sun, Wenhao Wang, Qiang Chen, Errui Ding, Yi Yang, and Jingdong Wang. 2024. MS-DETR: Efficient DETR Training with Mixed Supervision. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 17027–17036

    2024

  47. [55]

    Dehua Zheng, Wenhui Dong, Hailin Hu, Xinghao Chen, and Yunhe Wang. 2023. Less is More: Focus Attention for Efficient DETR. In Proceedings of the IEEE/CVF International Conference on Computer Vision . 6674–6683

    2023

  48. [56]

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

    2020 arXiv

  49. [57]

    flexible object predictions

    Zhuofan Zong, Guanglu Song, and Yu Liu. 2023. Detrs with collaborative hybrid assignments training. In Proceedings of the IEEE/CVF international conference on computer vision. 6748–6758. DS-Det: Single-Query Paradigm and Attention Disentangled Learning for Flexible Object Dete...

    2023

  50. [61]

    query ambiguity

    is achieved with the same weight as DINO. To improve the one-to-one classification accuracy of the model, we reduce the weight of the L1 loss in the DP component (index 1) and increase the weight of the classification loss (index 2). Building upon the baseline, we further incr...

    2025

  51. [900]

    query ambiguity

    while achieving a higher performance with a +0.9% AP, suggesting that the queries selected via FLET module are more effective. In terms of complexity, we have discussed the changes in FLOPs with varying queries in Sec. A.3. As shown in Fig. 6, the reduction in the number of qu...

    2025

  52. [2021]

    In Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Efficient attention: Attention with linear complexities. In Proceedings of the IEEE/CVF winter conference on applications of computer vision . 3531–3539

  53. [2022]

    arXiv preprint arXiv:2212.06137 (2022)

    Nms strikes back. arXiv preprint arXiv:2212.06137 (2022)

    2022 arXiv

  54. [2023]

    arXiv preprint arXiv:2304.07527 (2023)

    Align-DETR: Improving DETR with simple IoU-aware BCE loss. arXiv preprint arXiv:2304.07527 (2023)

    2023 arXiv

Pith tools

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