Pith. sign in

REVIEW 2 major objections 7 minor 2 cited by

LP-DETR: Layer-wise Progressive Relations for Object Detection

T0 review · 2 major / 7 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read LP-DETR claims that adding learnable multi-scale spatial relations to decoder self-attention, with the relation mix shifting from local to global across layers, improves DETR-based object detection.

desk verdict A clean, clearly specified incremental extension of Relation-DETR, but the headline accuracy gains are single-run numbers and could easily sit inside run-to-run noise. read the letter →

arxiv 2502.05147 v3 pith:RE73QXLY submitted 2025-02-07 cs.CV cs.AI

Zhengjian Kang , Ye Zhang , Xiaoyu Deng , Xintao Li , Yongzhe Zhang This is my paper · ORCID
classification cs.CVcs.AI
keywords objectdetectionDETRtransformerdecoderself-attentionspatialrelationsmulti-scalelayer-wiseprogressiveCOCO
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that DETR-style object detectors leave a useful signal unused: the spatial relations between object queries inside decoder self-attention. The authors propose LP-DETR, which augments each query with a relation-aware attention term built from geometric relation metrics — log relative distances, scale ratios, and IoU between reference boxes — at three scales: local, medium, and global. A learnable per-layer weight balances the three scales, and the medium scale is deliberately broadened as the decoder gets deeper. On COCO 2017 the method reaches 52.3 AP with ResNet-50 at 12 epochs and 52.5 AP at 24 epochs, exceeding its direct predecessor Relation-DETR by 0.4–0.6 AP, and 58.1 AP with a Swin-L backbone. The authors further report that the learned weights shift from local-dominated early layers to medium/global-dominated deep layers, which they interpret as evidence that detection benefits from hierarchical spatial reasoning.

What carries the argument

The load-bearing object is the layer-wise progressive relation-aware self-attention module inserted into the DETR decoder's self-attention. For each object query it computes a relation query as a weighted sum of content queries, with weights that combine ordinary attention scores with a geometric weight; the geometric weight is produced by an MLP over sinusoidal embeddings of local, medium, and global relation metrics derived from reference-box coordinates (Eqs. 1–7). The progressive part is the medium relation metric, whose log-scale input is divided by $(1 + 2l/L)$ at decoder layer $l$, making relations broader in later layers, together with a learnable per-layer vector $\Lambda = [\lambda_{\text{local}}, \lambda_{\text{medium}}, \lambda_{\text{global}}]$ that rebalances the three scales. This object carries the argument because it is the only new component added to the baseline detector: the gains are attributed to it, and the observed local-to-global weight trajectory is read off from the learned $\Lambda$ values.

What would settle it

A reader could settle the claim by training LP-DETR with the medium-relation scaling factor held constant across all six decoder layers (and, as a second arm, reversed so early layers see broad relations and later layers see narrow ones) while keeping the relation-aware attention module unchanged. If AP stays at 52.3 or the layer-wise weight curve still runs local-to-global, the progressive schedule is not the cause of the gain; if AP drops toward the 51.1 baseline of relation-aware attention without progressive refinement, the schedule is load-bearing.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that explicitly modeling multi-scale spatial relationships between object queries in the decoder's self-attention improves both convergence speed and final detection accuracy, and that the model learns a layer-wise progression from local to global relations. The mechanism works by enriching each object query with a relation query: a weighted sum of all content queries, where the weights multiply ordinary attention scores by a geometric weight. The geometric weight is an MLP applied to a sinusoidal embedding of a five-dimensional relation metric (log distance ratios, log width/height ratios, and IoU of the reference boxes), evaluated under three relation scales. The local scale is the raw metric, the global scale uses constant weights, and the medium scale applies a factor of $(1 + 2l/L)$ that grows with decoder layer index $l$, so deeper layers see progressively broader spatial neighborhoods. Results on COCO 2017, CSD, and MSSD show consistent gains over Relation-DETR, and an ablation attributes about 0.9 AP to the relation mechanism and another 0.3 AP to the progressive refinement schedule.

Load-bearing premise

The load-bearing premise is that the rule that later decoder layers use progressively broader relations — not just the presence of multi-scale geometric relations themselves — is what improves detection; if a fixed or reversed schedule kept the same accuracy and weight pattern, the paper's central explanation would collapse.

Editorial extensions

If this is right

  • Replacing standard decoder self-attention with relation-aware self-attention raises AP from 51.1 to 52.0 on COCO 2017 with ResNet-50 at 12 epochs; adding the progressive refinement schedule raises it further to 52.3.
  • Allocating more attention heads to relation-aware computation improves accuracy monotonically: zero relation heads gives 51.1 AP, while eight relation heads give 52.3 AP.
  • The gain over Relation-DETR is consistent across schedules and backbones: +0.6 AP at 12 epochs, +0.4 AP at 24 epochs with ResNet-50, and +0.3 AP with Swin-L, and the method also leads on the CSD and MSSD datasets.
  • The learned relation weights show a stable pattern: early decoder layers favor local relations (weights above 0.9), while later layers shift toward medium and global relations, with medium/global weights overtaking local by the final layer.
  • The model's convergence curve stays above Relation-DETR's throughout training, consistent with the claim that spatial relation modeling speeds up DETR training.

Reading between the lines

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

  • In our reading, the reported local-to-global progression is partly imposed by construction: the medium-relation scaling factor $(1 + 2l/L)$ grows with layer index before any training, so the learned $\Lambda$ weights may simply be amplifying a schedule the architecture already pushes toward. A fair test would hold the schedule fixed or reverse it and check whether the pattern and the +0.3 AP from
  • Because the relation module lives entirely in decoder self-attention, it is orthogonal to denoising-based training (e.g., DN-DETR, DINO) and one-to-many matching strategies; combining LP-DETR with those could plausibly stack gains, but the paper does not test this.
  • The relation metric is defined over reference boxes, so the same mechanism could transfer to other set-prediction tasks that use DETR-style decoders — panoptic segmentation, multi-object tracking, or open-vocabulary detection — wherever box coordinates are available. The paper does not explore these.
  • If the accuracy gain is driven mainly by the geometric relation prior rather than by the layer-wise progression, a simpler fixed mixture of local, medium, and global relations might achieve most of the benefit; that hypothesis is testable by ablating $\Lambda$ to constant values.
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

2 major / 7 minor

Summary. LP-DETR proposes a modification to DETR-style decoder self-attention in which object queries are augmented with multi-scale geometric relation features. The method defines local, medium, and global relation metrics based on log-distances, aspect ratios, and IoU, combines them with per-layer learnable weights Λ, and introduces a layer-dependent scaling factor for the medium relation. Experiments on COCO 2017, CSD, and MSSD report improvements over Relation-DETR (e.g., 52.3 vs. 51.7 AP at 12 epochs with ResNet-50 and 58.1 vs. 57.8 AP with Swin-L), accompanied by ablations on the number of relation heads, the IoU variant, and the progressive refinement module, as well as a qualitative analysis claiming that learned relation weights progress from local to global across decoder layers.

Significance. The core idea of explicit multi-scale geometric relation modeling in DETR self-attention is a reasonable and potentially useful extension of Relation-DETR, and the paper gives a relatively clear module description with several ablations. The paper's strengths include isolating the relation module in Table 5, varying the number of relation heads in Fig. 2a, testing different IoU variants in Fig. 2b, and evaluating on three datasets. However, the reported gains are small (0.3–0.6 AP) and are presented without error bars, seed counts, or significance tests; moreover, the claimed local-to-global progression is partly imposed by the hand-chosen layer-dependent medium-relation scaling in Eq. (7). If the accuracy gains survive multi-seed evaluation and the progression is shown to be learned rather than prescribed, this would be a modest but solid incremental contribution. In its current form, the evidence does not yet support the headline claims.

major comments (2)
  1. [§4.2, Tables 1–5] Every claimed improvement over Relation-DETR (Table 1: +0.6 AP at 12 epochs and +0.4 AP at 24 epochs; Table 2: +0.3 AP; Tables 3–4: +0.4 and +0.6 AP; Table 5: +0.9 AP for the relation module and +0.3 AP for progressive refinement) is based on a single run with no standard deviations, seed counts, or significance tests. DETR-style detectors trained for 12 epochs typically exhibit run-to-run AP variation of 0.2–0.5 AP, so the reported gains are comparable to plausible noise. The central accuracy claim is therefore not yet supported; please provide mean and standard deviation over at least three seeds for the main comparisons and the key ablations, or otherwise demonstrate that the differences are reproducible.
  2. [§3.2, Eq. (7) and §4.4] The claimed local-to-global progression is not a purely learned phenomenon. The medium-relation metric is defined with the layer-dependent scaling factor (1+2l/L), so the input to the learned weights Λ changes systematically with decoder depth before any training. The paper does not ablate this hand-chosen schedule, for example by fixing the factor to 1, removing it, or using an inverted schedule. Without such an ablation, the claim that the model 'naturally learns to prioritize local spatial relations in early decoder layers' is at least partly an artifact of the definition, and the accuracy contribution of the progressive component in Table 5 (+0.3 AP) cannot be attributed to learned progression.
minor comments (7)
  1. [Abstract, Section 1, and Table 2] The Swin-L result is reported as 58.0 AP in the Abstract and Introduction but as 58.1 AP in Table 2; please reconcile the headline number.
  2. [Equations (1)–(7)] The subscript and superscript notation in the equations is difficult to parse (for example, Eq. (1) mixes q_i^c, q_i^p, and q_i^r with an unclear summation index), and the dimensions of the projection matrices W_q, W_k, W_v, and W_g are not stated. Please rewrite the notation cleanly and define all tensor shapes.
  3. [Section 2.1 and reference list] Relation-DETR is cited as [11] in the section text but appears as [10] in the reference list, where [11] is Salience-DETR; please correct the citation labels.
  4. [§4.2 and Conclusion] For ResNet-50 at 12 epochs, Table 1 shows AP_S decreasing from 36.1 (Relation-DETR) to 35.8 (LP-DETR), so the conclusion's statement of 'marginal improvements on small objects' is not supported by the reported numbers; please report and discuss this decrease.
  5. [§4.4 and Fig. 3a] The analysis of local-to-global relation weights should state how the plotted values are normalized and what the y-axis represents; without this, statements such as 'weights exceeding 0.9' and 'decrease to 0.24' cannot be verified.
  6. [§4.2, Table 2] The statement that LP-DETR achieves 'state-of-the-art performance of 58.1 AP' is based on only four Swin-L baselines; please either expand the comparison to a broader set of recent Swin-L DETR detectors or qualify the claim as improvement over the listed baselines.
  7. [§4.1] CSD and MSSD are described as having 373 and 962 training images, respectively; please clarify whether these are the complete dataset sizes and specify the evaluation splits, since the small scale of these datasets makes single-run comparisons especially noisy.

Circularity Check

1 steps flagged · score 4.0 of 10

LP-DETR's detection gains are empirical, but its headline 'local-to-global progression' is partly built into the definition of the medium relation metric, making that interpretive claim partially circular.

  1. self definitional [Section 3.2, Eq. 7; Section 4.4, Fig. 3a]
    "The medium relation R m applies a scaling factor of (1+2×l/L) to reduce the log function's steepness... local relation weights decrease to 0.24, while medium and global relation weights steadily increase to around 0.4."

    The claimed discovered pattern—local relations dominating early layers and global relations growing deeper—is not purely learned. Equation 7 feeds the medium relation R_m into the geometric weight, and R_m is defined with the layer-dependent factor (1+2l/L), so the medium-relation log distances are deliberately flattened and made more global-like at deeper layers by construction. The observed shift in Fig. 3a is therefore a readout of a hand-set schedule combined with the fitted weights Λ, rather than an emergent property of the model. The paper provides no ablation where R_m is kept fixed across layers, so the 'natural progression' claim is partly an artifact of the metric definition. The detection accuracy numbers remain genuine empirical results, so the circularity is only partial.

full rationale

The central detection-accuracy claim (52.3 AP vs 51.7 AP, 58.1 vs 57.8) is an empirical outcome, not a fitted prediction, and the paper does not derive those numbers from its own assumptions. There are no load-bearing self-citations, uniqueness imports, or renamed known results. The one circular element is the paper's third contribution: that the model 'naturally learns to prioritize local spatial relations in early decoder layers while gradually shifting attention to broader contexts in deeper layers.' That pattern is substantially shaped by Eq. 7's definition of R_m with the layer-dependent scaling factor (1+2l/L), which makes the medium relation progressively blunter and more global-like at deeper layers before learning occurs. The learned weights in Fig. 3a are real, but the claimed 'natural' local-to-global progression is partly built into the input representation, and the paper does not test a fixed-scale variant. Because the accuracy improvement itself is not circular and the progressive-pattern claim is only one component of the paper, a moderate score of 4 is appropriate.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

This ledger counts the learnable per-layer weights Lambda as the main free parameter; the medium-relation scaling factor and the CIoU selection are hand-chosen hyperparameters. The axioms are standard relation-network and evaluation assumptions. The only invented construct is the relation query vector, which has no independent falsifiable handle outside the model.

free parameters (3)
  • Per-layer relation scale weights Lambda = Not reported as numbers; Fig. 3a shows normalized weights, e.g., layer 2 local >0.9, final layer local ~0.24…
    These learnable weights in Eq. 7 rebalance local, medium, and global relations in each decoder layer; they are fitted on COCO and are the core of the progressive relation claim.
  • Medium-relation scaling factor (1+2l/L) = Hand-chosen; no search reported
    Defined in Section 3.2, this layer-dependent schedule encodes the local-to-global broadening into the medium metric and is not learned or ablated.
  • IoU variant choice (CIoU) = CIoU selected after comparing with IoU, GIoU, DIoU
    Section 4.3 and Fig. 2b; chosen for marginally superior results, an empirical selection on validation data.
assumptions (3)
  • domain assumption Geometric relations between predicted reference boxes provide useful inductive bias for query interaction in DETR decoder self-attention.
    Adopted from Relation-DETR [10] and Relation Networks [12]; the paper builds on this without re-deriving it. Invoked in Section 3.2.
  • standard math Sinusoidal position encoding followed by a learnable MLP can embed the 5-D geometric relation feature into a high-dimensional space for attention weighting.
    Standard practice from Transformers [28] and Relation Networks; used in Eqs. 4 to 6 without proof. It is a background assumption, not ad hoc to this paper specifically.
  • domain assumption The standard COCO evaluation protocol with a single training run is sufficient to compare detection methods.
    The paper reports one-run AP values with no error bars; this assumes differences of 0.3 to 0.6 AP are meaningful, which is conventional but not statistically verified.
invented entities (1)
  • Relation query q_i^r
    purpose: A query-level representation formed as the weighted sum of all object queries, carrying geometric and attention-based relation information, then concatenated with the original query (Eqs. 1 to 6).
    An internal network component with no external falsifiable prediction; its effect is only measurable indirectly through downstream AP. It is a modeling construct, not a new scientific entity with independent evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LP-DETR: Layer-wise Progressive Relations for Object Detection." pith.science (2026). https://pith.science/paper/RE73QXLY

@misc{pith2026250205147,
  author       = {Pith},
  title        = {Pith review of: LP-DETR: Layer-wise Progressive Relations for Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RE73QXLY}},
  note         = {Machine review of arXiv:2502.05147}
}
read the original abstract

This paper presents LP-DETR (Layer-wise Progressive DETR), a novel approach that enhances DETR-based object detection through multi-scale relation modeling. Our method introduces learnable spatial relationships between object queries through a relation-aware self-attention mechanism, which adaptively learns to balance different scales of relations (local, medium and global) across decoder layers. This progressive design enables the model to effectively capture evolving spatial dependencies throughout the detection pipeline. Extensive experiments on COCO 2017 dataset demonstrate that our method improves both convergence speed and detection accuracy compared to standard self-attention module. The proposed method achieves competitive results, reaching 52.3\% AP with 12 epochs and 52.5\% AP with 24 epochs using ResNet-50 backbone, and further improving to 58.0\% AP with Swin-L backbone. Furthermore, our analysis reveals an interesting pattern: the model naturally learns to prioritize local spatial relations in early decoder layers while gradually shifting attention to broader contexts in deeper layers, providing valuable insights for future research in object detection.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Predicting Asphalt Pavement Friction Using Texture-Based Image Indicator

    cs.CV 2025-07 conditional novelty 4.0 of 10

    The paper defines aggregate protrusion area from 2D pavement images as a texture indicator and reports linear models with adjusted R2 above 0.90 for predicting DFT friction for three asphalt surface types.

  2. DK-RRT: Deep Koopman RRT for Collision-Aware Motion Planning of Space Manipulators in Dynamic Debris Environments

    cs.RO 2025-07 reject novelty 3.0 of 10

    DK-RRT is described as a deep Koopman-based extension of RRT for space manipulators, but the paper lacks any actual RRT implementation and its simulation baselines are undefined.

Reference graph

Works this paper leans on

46 extracted references · 45 canonical work pages · cited by 2 Pith papers

  1. [1]

    This paper presents LP-DETR (Layer-wise Progressive Relation DETR), a novel approach that enhances DETR-based object detection through multi-scale relation modeling

    LP-DETR: Layer-wise Progressive Relation for Object Detection Zhengjian Kang1*, Ye Zhang2*, Xiaoyu Deng3, Xintao Li4, Yongzhe Zhang5 1 New York University, NY 10012, USA 2 University of Pittsburgh, PA 15213, USA 3 Fordham University, NY 10458, USA 4 Georgia Institute of Technology, GA 30332, USA 5 California Institute of Technology, CA 91125, USA 1zk299@n...

    work page 2017

  2. [2]

    Method Backbone Epochs AP AP_50 AP_75 AP_S AP_M AP_L DINO

    Evaluation on COCO 2017 with SOTA methods using Swin-L. Method Backbone Epochs AP AP_50 AP_75 AP_S AP_M AP_L DINO

    work page 2017

  3. [3]

    =#𝑤!#"$ #⋅&𝑊%⋅𝑞#( (1) The relation weight 𝑤'-* captures both geometric and attention-based relationships be-tween queries, which is computed as: 𝑤!#

    Our proposed progressive relation-aware attention is applied into the self-attention in the decoder component. Let's consider object queries 𝐐 consists of label embedding queries 𝐐𝐜, reference box position queries 𝐐𝐩 (represented by (𝑥,𝑦,𝑤,ℎ)), Backbone Transformer Encoder Encoder memory Input image tokenize learnable object queries Relation-aware Self-At...

    work page 2017

  4. [4]

    et al.: Group DETR: Fast DETR Training with Group-wise One-to-many Assign-ment

    Chen, Q. et al.: Group DETR: Fast DETR Training with Group-wise One-to-many Assign-ment. In: Intl. Conf. on Computer Vision (ICCV). pp. 6633–6642 (2023)

    work page 2023

  5. [5]

    et al.: Iterative Visual Reasoning Beyond Convolutions

    Chen, X. et al.: Iterative Visual Reasoning Beyond Convolutions. In: IEEE Conf. on Com-puter Vision and Pattern Recognition (CVPR). pp. 7239–7248 (2018)

    work page 2018

  6. [6]

    COCO 2017 contains 113k train-ing images across 80 object categories

    datasets. COCO 2017 contains 113k train-ing images across 80 object categories. CSD and MSSD are task-specific datasets in-cluding 373 training images with 3 categories and 962 training images with 4 catego-ries, respectively. The performance is evaluated using standard COCO metrics: average precision (AP) at different IoU thresholds (IoU=0.5, 0.75, 0.5:0...

    work page 2017

  7. [7]

    Method Backbone Epochs AP AP_50 AP_75 AP_S AP_M AP_L Def-DETR

    Evaluation on COCO 2017 with SOTA methods using ResNet-50. Method Backbone Epochs AP AP_50 AP_75 AP_S AP_M AP_L Def-DETR

    work page 2017

  8. [8]

    et al.: Relation-enhanced DETR for Component Detection in Graphic Design Re-verse Engineering

    Hao, X. et al.: Relation-enhanced DETR for Component Detection in Graphic Design Re-verse Engineering. In: Intl. Joint Conference on Artificial Intelligence (IJCAI). pp. 4785–4793 (2023)

    work page 2023

Show all 46 references
  1. [9]

    et al.: SRRV: A Novel Document Object Detector Based on Spatial-Related Relation and Vision

    Bi, H. et al.: SRRV: A Novel Document Object Detector Based on Spatial-Related Relation and Vision. IEEE Trans. Multimed. 25, 3788–3798 (2023)

    2023

  2. [10]

    introduces explicit position relations be-tween bounding boxes with cross-layer refinement. Motivated by their work, but dif-ferent from these approaches, we directly incorporate geometric relation weights into queries within each layer and propose layer-wise progressive relat...

    2017

  3. [11]

    et al.: Align-DETR: Improving DETR with Simple IoU-aware BCE Loss

    Cai, Z. et al.: Align-DETR: Improving DETR with Simple IoU-aware BCE Loss. ArXiv Prepr. ArXiv230407527. (2023)

    2023

  4. [12]

    et al.: End-to-End Object Detection with Transformers

    Carion, N. et al.: End-to-End Object Detection with Transformers. In: European Conf. on Computer Vision (ECCV). pp. 213–229 Springer (2020)

    2020

  5. [13]

    et al.: DAC-DETR: Divide the Attention Layers and Conquer

    Hu, Z. et al.: DAC-DETR: Divide the Attention Layers and Conquer. Adv. Neural Inf. Pro-cess. Syst. 36, (2024)

    2024

  6. [14]

    et al.: EEE-Net: Efficient Edge Enhanced Network for Surface Defect Detection of Glass

    Chen, Y. et al.: EEE-Net: Efficient Edge Enhanced Network for Surface Defect Detection of Glass. IEEE Trans Instrum. Meas. (2023)

    2023

  7. [15]

    et al.: ImageNet: A Large-scale Hierarchical Image Database

    Deng, J. et al.: ImageNet: A Large-scale Hierarchical Image Database. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 248–255 IEEE (2009)

    2009

  8. [16]

    Position re-lations are constructed at three scales (local, medium, and global) with equal initial weights on 6 decoder layers

    For the relation embedding module, we set the temperature 𝑇=10000, scale 𝑠=100, and position embedding dimension 𝑑pos=16 in sinusoidal encoding. Position re-lations are constructed at three scales (local, medium, and global) with equal initial weights on 6 decoder layers. Foll...

    2017

  9. [17]

    et al.: Deep Residual Learning for Image Recognition

    He, K. et al.: Deep Residual Learning for Image Recognition. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016)

    2016

  10. [18]

    et al.: Relation DETR: Exploring Explicit Position Relation Prior for Object De-tection

    Hou, X. et al.: Relation DETR: Exploring Explicit Position Relation Prior for Object De-tection. In: European conf. on computer vision. Springer (2024)

    2024

  11. [19]

    et al.: Salience DETR: Enhancing Detection Transformer with Hierarchical Sali-ence Filtering Refinement

    Hou, X. et al.: Salience DETR: Enhancing Detection Transformer with Hierarchical Sali-ence Filtering Refinement. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 17574–17583 (2024)

    2024

  12. [20]

    et al.: Relation Networks for Object Detection

    Hu, H. et al.: Relation Networks for Object Detection. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 3588–3597 (2018)

    2018

  13. [21]

    et al.: Microsoft COCO: Common Objects in Context

    Lin, T.-Y. et al.: Microsoft COCO: Common Objects in Context. In: European Conf. on Computer Vision (ECCV). pp. 740–755 Springer (2014)

    2014

  14. [22]

    et al.: DETRs with Hybrid Matching

    Jia, D. et al.: DETRs with Hybrid Matching. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 19702–19712 (2023)

    2023

  15. [23]

    Kingma, D.P., Ba, J.: Adam: A Method for Stochastic Optimization, (2017)

    2017

  16. [24]

    et al.: Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations

    Krishna, R. et al.: Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations. Intl J. Comput. Vis. 123, 32–73 (2017)

    2017

  17. [25]

    et al.: DN-DETR: Accelerate DETR Training by Introducing Query Denoising

    Li, F. et al.: DN-DETR: Accelerate DETR Training by Introducing Query Denoising. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 13619–13627 (2022)

    2022

  18. [26]

    et al.: GAR: Graph Assisted Reasoning for Object Detection

    Li, Z. et al.: GAR: Graph Assisted Reasoning for Object Detection. In: Proceedings of the IEEE/CVF Winter Conf. on Applications of Computer Vision. pp. 1295–1304 (2020)

    2020

  19. [27]

    et al.: Revolutionizing Drug Discovery: Integrating Spatial Transcriptomics with Ad-vanced Computer Vision Techniques

    Li, Z. et al.: Revolutionizing Drug Discovery: Integrating Spatial Transcriptomics with Ad-vanced Computer Vision Techniques. In: 1st CVPR Workshop on Computer Vision For Drug Discovery (CVDD): Where are we and What is Beyond? (2025)

    2025

  20. [28]

    et al.: Core-Text: Improving Scene Text Detection with Contrastive Relational Rea-soning

    Lin, J. et al.: Core-Text: Improving Scene Text Detection with Contrastive Relational Rea-soning. In: Intl. Conf. on Multimedia and Expo (ICME). pp. 1–6 IEEE (2021). 12 Z. Kang et al

    2021

  21. [29]

    et al.: A Casting Surface Dataset and Benchmark for Subtle and Confusable De-fect Detection in Complex Contexts

    Wang, Q. et al.: A Casting Surface Dataset and Benchmark for Subtle and Confusable De-fect Detection in Complex Contexts. IEEE Sens. J. 1–1 (2024)

    2024

  22. [30]

    et al.: DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR

    Liu, S. et al.: DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR. In: Intl. Conf. on Learning Representations (ICLR). (2021)

    2021

  23. [31]

    et al.: Detection Transformer with Stable Matching

    Liu, S. et al.: Detection Transformer with Stable Matching. ArXiv Prepr. ArXiv230404742. (2023)

    2023

  24. [32]

    et al.: Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows

    Liu, Z. et al.: Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows. In: Intl. Conf. on Computer Vision (ICCV). pp. 10012–10022 (2021)

    2021

  25. [33]

    et al.: Rank-DETR for High Quality Object Detection

    Pu, Y. et al.: Rank-DETR for High Quality Object Detection. Adv. Neural Inf. Process. Syst. 36, (2024)

    2024

  26. [34]

    et al.: Factors influencing long-term city park visitations for mid-sized US cities: A big data study using smartphone user mobility

    Song, Y. et al.: Factors influencing long-term city park visitations for mid-sized US cities: A big data study using smartphone user mobility. Sustain. Cities Soc. 80, 103815 (2022)

    2022

  27. [35]

    Tianqi et al.: NeRF-Based defect detection, http://arxiv.org/abs/2504.00270, (2025)

    2025 arXiv

  28. [36]

    et al.: Attention is All You Need

    Vaswani, A. et al.: Attention is All You Need. Adv. Neural Inf. Process. Syst. 30, (2017)

    2017

  29. [37]

    et al.: Deformable DETR: Deformable Transformers for End-to-end Object Detec-tion

    Zhu, X. et al.: Deformable DETR: Deformable Transformers for End-to-end Object Detec-tion. In: Intl. Conf. on Learning Representations (ICLR). (2020)

    2020

  30. [38]

    et al.: Reasoning-RCNN: Unifying Adaptive Global Reasoning into Large-scale Ob-ject Detection

    Xu, H. et al.: Reasoning-RCNN: Unifying Adaptive Global Reasoning into Large-scale Ob-ject Detection. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 6419–6428 (2019)

    2019

  31. [39]

    et al.: Cascade-DETR: Delving into High-quality Universal Object Detection

    Ye, M. et al.: Cascade-DETR: Delving into High-quality Universal Object Detection. In: Intl. Conf. on Computer Vision (ICCV). pp. 6704–6714 (2023)

    2023

  32. [40]

    et al.: DINO: DETR with Improved Denoising Anchor Boxes for End-to-end Object Detection

    Zhang, H. et al.: DINO: DETR with Improved Denoising Anchor Boxes for End-to-end Object Detection. In: Intl. Conf. on Learning Representations (ICLR). (2022)

    2022

  33. [41]

    et al.: Dense Distinct Query for End-to-end Object Detection

    Zhang, S. et al.: Dense Distinct Query for End-to-end Object Detection. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 7329–7338 (2023)

    2023

  34. [42]

    et al.: MS-DETR: Efficient DETR Training with Mixed Supervision

    Zhao, C. et al.: MS-DETR: Efficient DETR Training with Mixed Supervision. In: IEEE Conf. on Computer Vision and Pattern Recognition (CVPR). pp. 17027–17036 (2024)

    2024

  35. [43]

    et al.: Transformer-Based Dual Relation Graph for Multi-Label Image Recognition

    Zhao, J. et al.: Transformer-Based Dual Relation Graph for Multi-Label Image Recognition. In: Proceedings of the IEEE/CVF Intl. Conf. on Computer Vision (ICCV). pp. 163–172 (2021)

    2021

  36. [44]

    et al.: Less is More: Focus Attention for Efficient DETR

    Zheng, D. et al.: Less is More: Focus Attention for Efficient DETR. In: 2023 IEEE/CVF Intl. Conf. on Computer Vision (ICCV). pp. 6651–6660 (2023)

    2023

  37. [46]

    et al.: DETRs with Collaborative Hybrid Assignments Training

    Zong, Z. et al.: DETRs with Collaborative Hybrid Assignments Training. In: Intl. Conf. on Computer Vision (ICCV). pp. 6748–6758 (2023)

    2023

  38. [2017]

    The results are summarized in Table 1 and Table

    We evaluate our model on the COCO 2017 validation dataset. The results are summarized in Table 1 and Table

    2017

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.