Pith. sign in

REVIEW 5 major objections 6 minor 2 cited by

YOLO-Master claims a new accuracy-latency Pareto point for nano-scale real-time detection by replacing static dense computation with a sparse mixture-of-experts block that routes each input to its most relevant experts.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 13:39 UTC pith:3N7RFUYD

load-bearing objection Incremental MoE-on-YOLO paper where the speedup claim likely evaporates under a controlled latency benchmark; the accuracy gain is small but plausible, and the ablations are worth a look. the 5 major comments →

arxiv 2512.23273 v2 pith:3N7RFUYD submitted 2025-12-29 cs.CV

YOLO-Master: MOE-Accelerated with Specialized Transformers for Enhanced Real-time Detection

Xu Lin , Jinlong Peng , Zhenye Gan , Jiawen Zhu , Jun Liu This is my paper
classification cs.CV
keywords real-time object detectionmixture of expertsconditional computationYOLOsparse routingload balancing lossdepthwise separable convolutiondynamic routing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

YOLO-Master claims that the fixed computational budget of YOLO-style real-time detectors is the main thing holding back their accuracy-efficiency trade-off, and that a sparse mixture-of-experts block can fix it by routing each input to a small set of specialized experts. Its Efficient Sparse MoE (ES-MoE) block uses a global-average-pooled routing descriptor to activate only the top-2 of four depthwise-separable-convolution experts at inference, after training with soft top-K routing and a load-balancing loss. On MS COCO the nano model reaches 42.4% AP at 1.62 ms, beating YOLOv13-N by 0.8 mAP while running 17.8% faster, with the biggest gains on dense small-object benchmarks like VisDrone. The paper also shows that the ES-MoE block belongs in the backbone, not the neck, and that a MoE-only loss beats a DFL+MoE combination. If the claim holds, conditional computation becomes a practical lever for real-time detection rather than a large-language-model technique.

Core claim

On its own terms, the paper claims that a real-time object detector can escape the static accuracy-latency trade-off by making computation instance-conditional. The ES-MoE block pools the input feature map into a single global descriptor, computes a softmax weight for each expert through two 1x1 convolutions, and at inference time executes only the top-2 of four depthwise-separable-convolution experts (kernel sizes 3, 5, 7). Training uses soft top-K to preserve gradients and a load-balancing MSE loss to keep expert utilization uniform. With this block in the backbone only, YOLO-Master-N reports 42.4% AP at 1.62 ms on COCO, outperforming YOLOv13-N by 0.8 mAP while being 17.8% faster, and the

What carries the argument

The load-bearing object is the Efficient Sparse Mixture-of-Experts (ES-MoE) block: a dynamic routing network that reduces the input feature map to a global descriptor via global average pooling, passes it through two 1x1 convolutions (channel reduction to C/8), and produces E softmax logits. During inference only the top-K experts—depthwise separable convolutions with kernel sizes 3, 5, and 7—are executed (hard top-K), while training uses soft top-K with a renormalized mask; an MSE load-balancing loss keeps all experts used. The block is inserted into the backbone feature extractor, and the paper identifies backbone-only placement as essential, with neck or full integration degrading perform

Load-bearing premise

The routing decision is made from a single global-average-pooled descriptor of the entire feature map and then applied uniformly to every spatial location, so the premise that the model 'allocates resources according to scene complexity' rests on the assumption that a global pooling signal preserves enough spatial information about where extra capacity is needed.

What would settle it

Ablate the GAP by feeding the unpooled feature map directly to the routing network (or by using a spatial or regional router) while keeping all other settings identical; if dense-scene AP does not improve, the global-descriptor premise is not the limiting factor, but if it does, the paper's core mechanism is incomplete. Alternatively, log the per-image expert choices and check that complex scenes systematically activate a different expert subset than simple scenes.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the reported numbers hold, YOLO-Master-N establishes a new Pareto point for nano-scale real-time detection: 42.4% AP at 1.62 ms, surpassing YOLOv13-N by 0.8 mAP with 17.8% lower latency.
  • The consistent gains on dense benchmarks (VisDrone +2.1 mAP, KITTI +1.5 mAP) indicate that sparse expert routing helps exactly where static detectors are argued to under-allocate capacity—scenes with many small objects.
  • The ablation showing that backbone-only ES-MoE improves over baseline (+1.3 mAP) while neck-only and full integration hurt (-2.6 and -5.9 mAP) implies a design rule: MoE routing should sit at feature extraction, not fusion, and cascaded routers can destabilize training.
  • The result that a MoE-only loss (62.2 mAP) outperforms DFL+MoE (61.9 mAP) suggests that for this architecture the load-balancing loss subsumes the distribution focal loss, a potentially useful simplification for future MoE-based detectors.
  • The same backbone transfers to classification (76.6% ImageNet top-1, +4.9 over YOLOv12-N) and instance segmentation (35.6 mask mAP, +2.8 over YOLOv12-seg-N), so the benefit is not confined to the detection head.

Where Pith is reading between the lines

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

  • An untested extension suggested by the design: replacing the global-average-pooled routing vector with per-region or per-cell gating could sharpen the adaptive allocation, since the current router applies one expert choice to the whole feature map and cannot distribute capacity within a mixed-complexity scene.
  • The paper does not ablate the routing signal itself (e.g., GAP versus spatial attention or token-wise routing), so the 'instance-conditional' claim is not yet distinguished from a simple per-image expert preference; comparing these variants would isolate the source of the gains.
  • The 17.8% speedup is reported on one accelerator at batch size 1; the actual latency benefit of depthwise-separable experts and sparse activation is hardware-dependent, so the Pareto improvement may shrink or grow on mobile or edge chips.
  • Since the backward pass still touches all experts via soft top-K weights, the training cost does not reflect inference sparsity; a fully sparse training objective would be the logical next step if training latency matters.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes YOLO-Master, a YOLO-family real-time detector that inserts Efficient Sparse Mixture-of-Experts (ES-MoE) blocks into the backbone and, in the general framework, the neck. A global average-pooling-based gating network produces per-image expert weights; soft top-K routing is used during training and hard top-K at inference. A load-balancing MSE loss is added to the YOLOv8 detection loss. The authors report state-of-the-art nano-scale results on five benchmarks, including 42.4% AP on COCO at 1.62 ms latency, outperforming YOLOv13-N by +0.8 AP and 17.8% faster inference. They also provide ablations on ES-MoE placement, number of experts, top-K, and loss configuration, plus classification and segmentation extensions.

Significance. If the reported Pareto improvement survives controlled benchmarking, the paper would be a useful contribution: it demonstrates a practical way to add conditional computation to a lightweight CNN detector, with a sensible soft/hard routing split and a load-balancing loss. Strengths include the breadth of ablations (placement, E, K, loss), evaluation on five datasets, and the stated intention to release code. However, the current evidence is weakened by uncontrolled latency measurements, absence of error bars or multi-seed results, post-hoc selection of the final configuration, and under-specified cross-task experiments. The contribution is promising but not yet established to the standard expected for a journal publication.

major comments (5)
  1. [Section 4.1 / Table 1 / Figure 1] Section 4.1 states that latency is measured on 'a dedicated inference accelerator' following 'the standard hardware configuration of the YOLOv12 baseline (FP16, batch size=1)', but it gives no accelerator model, inference framework (TensorRT, PyTorch, etc.), or measurement protocol (warmup, repetitions). The baseline latencies in Table 1 (e.g., YOLOv11-N: 1.50 ms, YOLOv13-N: 1.97 ms) appear to be taken from the original papers, which may have used different hardware/software stacks. Because the central claim is an accuracy-latency Pareto improvement (Figure 1), the 17.8% speed advantage over YOLOv13-N is load-bearing and could be an artifact of the benchmark environment rather than the architecture. Please report controlled measurements on the same hardware and framework, with error bars, or soften the speed claim accordingly.
  2. [Section 4.3 / Table 8 / Figure 3] No multiple seeds or error bars are reported for any accuracy number. The final configuration (Config 5, MoE-only loss with λ=1.5) was selected by comparing validation mAP across configurations in the ablation, and the same configuration is then presented as the final model in Table 1. Some of the claimed gains (e.g., +0.8 AP on COCO, +1.4 AP on VOC) are small relative to typical seed-to-seed variation for 600-epoch training. Please provide mean±std over at least three seeds, or otherwise justify that the reported deltas exceed noise. Also clarify whether the numbers in Table 1 are on the validation split or a held-out test split.
  3. [Table 5 / Section 4.3.1] Full integration of ES-MoE in both backbone and neck collapses mAP by 5.9 points (60.8→54.9). The paper attributes this to 'gradient interference between cascaded routing mechanisms' but provides no gradient analysis or diagnostic evidence. This is not a secondary finding: the framework diagram in Figure 2 and the text in Section 3.1 describe ES-MoE in both backbone and neck, while the deployed model is backbone-only. The architecture as advertised is thus not the architecture evaluated. The authors should either analyze the failure (e.g., gradient norms, routing entropy) and propose a fix, or restrict the paper's architectural claims to the backbone-only configuration.
  4. [Section 3.3 / Eq. (4)-(5)] The gating network compresses the entire feature map with global average pooling and produces a single E-dimensional softmax vector applied to all spatial locations (Eq. 4). Yet Section 3.1 claims the design 'enables dynamic allocation of computational resources based on the local characteristics and complexity of input features.' A global vector cannot adapt to local spatial complexity; the mechanism is image-conditional, not spatially adaptive. No ablation compares this global routing with spatial or token-wise routing, so there is no evidence that global average pooling preserves the local information needed for the stated adaptive allocation. Please either add such an ablation or revise the 'local characteristics' claim.
  5. [Section 4.3.5 / Tables 3-4] The classification and segmentation experiments are reported without any training setup (epochs, resolution, augmentation, optimizer, loss). The ImageNet Top-1 gain of 76.6 vs. 71.7 over YOLOv12-cls-N is surprisingly large and, without a reproducible recipe, is not credible. These results are used to claim that YOLO-Master is a 'powerful and general-purpose architecture,' so they should be either fully specified or removed.
minor comments (6)
  1. [Section 3.2] The sentence about multi-kernel approaches contains an unresolved citation: 'Inspired by multi-kernel approaches in Inception networks [?]'. Please add the missing reference.
  2. [Abstract] There is stray text at the end of the abstract: '/githubCode: isLinXu/YOLO-Master'. This appears to be a leftover from the submission template and should be removed.
  3. [Figure 4 caption] Typo: 'MS COCO [24] ane PASCAL VOC' should be 'and PASCAL VOC'. Also, 'YOLOV10-N' is inconsistently capitalized.
  4. [Section 4.1] The sentence 'All training and testing are performed on 4 high-performance compute' is incomplete. Specify the hardware and relevant library versions.
  5. [Equations (4) and (6)] Notation for the routing weights is inconsistent: Eq. (4) uses ω_i, while Eq. (6) uses Ω. Please unify.
  6. [Table 1] Several baselines share identical SKU-110K mAP50 values (90.0). Please verify that these are not copied or rounded values; if they are identical by chance, state so.

Circularity Check

0 steps flagged

No significant circularity: the reported accuracy-latency claims are benchmarked externally and no prediction reduces to a fitted input or self-citation by construction.

full rationale

The derivation chain is architectural: the ES-MoE block is defined by Eqs. (1)-(5), phased routing by Eqs. (6)-(10), and the total loss by Eqs. (11)-(13). None of these equations define, or are fitted to, the headline COCO AP (42.4%) or latency (1.62 ms); those are instead external benchmark measurements reported against YOLOv10-13. The MoE routing and load-balancing loss follow standard literature cited to independent external works [7,14,35], and the few self-citations ([28], [39]) are background applications, not load-bearing premises. The main empirical risk—selection of Config 5 (MoE-only loss, λ=1.5) via validation mAP in Table 8, and later reporting that configuration's benchmark numbers—is validation-set model selection, a statistical overfitting concern, not a circular reduction: the final mAP is not equal to the loss weight by construction. Likewise, the uncontrolled latency comparison (Section 4.1) is a benchmarking-validity concern, not circularity. No passage in the manuscript asserts that a missing proof or unsupported premise is required for the central claim. Thus, under the requirement to exhibit a specific reduction before flagging circularity, the paper shows no significant circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities; the 'ES-MoE block' is a composition of standard operations. The main ledger entries are hyperparameters selected post hoc and assumptions about the sufficiency of global routing and real-hardware sparsity.

free parameters (5)
  • Number of experts E = 4
    Chosen via ablation (Table 6); 2 experts lose 1.3% mAP, 8 experts add parameters without gain.
  • Top-K = 2
    Chosen via ablation (Table 7); K=1 loses 0.5% mAP, K=3,4 give no gain.
  • Load-balancing loss weight λ_LB = 1.5
    Chosen via loss ablation (Table 8, Config 5); other weights and DFL inclusion show lower or unstable mAP.
  • Channel reduction ratio γ = 8
    Set by hand to keep gating network light; no ablation reported.
  • Expert kernel sizes = {3,5,7}
    Hand-selected odd sizes to cover receptive fields; no ablation reported.
axioms (5)
  • domain assumption YOLOv8-style detection loss (cls + loc + DFL) is an appropriate optimization target; the final model drops DFL and uses only MoE loss plus classification/localization.
    Section 3.5 uses L_YOLO with a claim that MoE loss subsumes DFL; this is an empirical assertion from one ablation run.
  • ad hoc to paper Global average pooling preserves enough information for expert selection.
    Section 3.3 derives routing logits solely from GAP of the entire feature map; no spatial routing or ablation is provided to justify this.
  • domain assumption Soft top-K at training transfers to hard top-K at inference without significant distribution shift.
    Section 3.4 switches between soft and hard top-K; the paper does not analyze calibration or domain shift.
  • domain assumption Hard top-K inference yields real speedup on target hardware.
    Section 4.1 reports latency on an unspecified accelerator; no profiling demonstrates that skipping experts reduces wall-clock time.
  • ad hoc to paper Depthwise separable convolutions with different kernel sizes are sufficient diversity for experts.
    Section 3.2 defines experts as DW-convs of sizes 3,5,7; no comparison with other expert architectures is given.

pith-pipeline@v1.3.0-alltime-deepseek · 13291 in / 12698 out tokens · 107662 ms · 2026-08-03T13:39:48.379868+00:00 · methodology

0 comments
read the original abstract

Existing Real-Time Object Detection (RTOD) methods commonly adopt YOLO-like architectures for their favorable trade-off between accuracy and speed. However, these models rely on static dense computation that applies uniform processing to all inputs, misallocating representational capacity and computational resources such as over-allocating on trivial scenes while under-serving complex ones. This mismatch results in both computational redundancy and suboptimal detection performance. To overcome this limitation, we propose YOLO-Master, a novel YOLO-like framework that introduces instance-conditional adaptive computation for RTOD. This is achieved through a Efficient Sparse Mixture-of-Experts (ES-MoE) block that dynamically allocates computational resources to each input according to its scene complexity. At its core, a lightweight dynamic routing network guides expert specialization during training through a diversity enhancing objective, encouraging complementary expertise among experts. Additionally, the routing network adaptively learns to activate only the most relevant experts, thereby improving detection performance while minimizing computational overhead during inference. Comprehensive experiments on five large-scale benchmarks demonstrate the superiority of YOLO-Master. On MS COCO, our model achieves 42.4% AP with 1.62ms latency, outperforming YOLOv13-N by +0.8% mAP and 17.8% faster inference. Notably, the gains are most pronounced on challenging dense scenes, while the model preserves efficiency on typical inputs and maintains real-time inference speed. Code will be available.

Figures

Figures reproduced from arXiv: 2512.23273 by Jiawen Zhu, Jinlong Peng, Jun Liu, Xu Lin, Zhenye Gan.

Figure 1
Figure 1. Figure 1: Accuracy-latency trade-off on MS COCO. YOLO [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The framework of YOLO-Master. The architecture integrates ES-MoE modules into the Backbone and Neck for enhanced feature [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Loss ablation study across configurations. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison across four challenging scenarios. All test images are from the MS COCO [ [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 2 Pith papers

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

  1. SARES-DEIM: Sparse Mixture-of-Experts Meets DETR for Robust SAR Ship Detection

    cs.CV 2026-04 unverdicted novelty 7.0

    SARES-DEIM achieves 76.4% mAP50:95 and 93.8% mAP50 on HRSID by routing SAR features through sparse frequency and wavelet experts plus a high-resolution preservation neck, outperforming prior YOLO and SAR detectors.

  2. CollabOD: Collaborative Multi-Backbone with Cross-scale Vision for UAV Small Object Detection

    cs.CV 2026-03 conditional novelty 4.0

    CollabOD improves UAV small-object detection via dual-path detail preservation, dense aggregation, bilateral reweighting, and a reparameterized detail-aware head, reporting 52.4 AP50 on VisDrone at 65.5 GFLOPs.

Reference graph

Works this paper leans on

45 extracted references · 11 linked inside Pith · cited by 2 Pith papers

  1. [1]

    End-to-end object detection with trans- formers.European Conference on Computer Vision, pages 213–229, 2020

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with trans- formers.European Conference on Computer Vision, pages 213–229, 2020

    2020

  2. [2]

    Modality-agnostic mixed-expert training for vision- language models

    Yi Chen, Hongchen Tan, Tianqi Wang, et al. Modality-agnostic mixed-expert training for vision- language models. InInternational Conference on Ma- chine Learning (ICML), 2023

    2023

  3. [3]

    Cswin transformer: A general vision transformer backbone with cross-shaped windows

    Xiaoyi Dong, Jianmin Bao, Dongdong Chen, et al. Cswin transformer: A general vision transformer backbone with cross-shaped windows. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12124–12134, 2022

    2022

  4. [4]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representations, 2021

    2021

  5. [5]

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

    Dawei Du, Pengfei Zhu, Longyin Wen, Xiao Bian, Haibin Lin, Qinghua Hu, Tao Peng, Jiayu Zheng, Xinyao Wang, Yue Zhang, Liefeng Bo, Hailin Shi, Rui Zhu, Aashish Kumar, Aijin Li, Almaz Zinollayev, Anuar Askergaliyev, Arne Schumann, Binjie Mao, Byeongwon Lee, Chang Liu, Changrui Chen, Chun- hong Pan, Chunlei Huo, Da Yu, DeChun Cong, Den- ing Zeng, Dheeraj Red...

    2019

  6. [6]

    Mark Everingham, Luc Van Gool, Christopher K. I. Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (VOC) challenge.Inter- national Journal of Computer Vision, 88(2):303–338, 2010

    2010

  7. [7]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.Journal of Machine Learning Research, 23(120):1–39, 2022

    William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.Journal of Machine Learning Research, 23(120):1–39, 2022

    2022

  8. [8]

    Are we ready for autonomous driving? the KITTI vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the KITTI vision benchmark suite. InConference on Computer Vision and Pattern Recognition (CVPR), 2012

    2012

  9. [9]

    Precise detection in densely packed scenes

    Eran Goldman, Roei Herzig, Avital Eisenschtat, Ja- cob Goldberger, and Tal Hassner. Precise detection in densely packed scenes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 3254–3263, 2019

    2019

  10. [10]

    Coordinate attention for efficient mobile network design.Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13713–13722, 2021

    Qibin Hou, Daquan Zhou, and Jiashi Feng. Coordinate attention for efficient mobile network design.Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13713–13722, 2021

    2021

  11. [11]

    Mobilenets: Efficient convolutional neural networks for mobile vision appli- cations.arXiv preprint arXiv:1704.04861, 2017

    Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. Mobilenets: Efficient convolutional neural networks for mobile vision appli- cations.arXiv preprint arXiv:1704.04861, 2017

    Pith/arXiv arXiv 2017

  12. [12]

    Gather-excite: Exploiting feature context in convolutional neural net- works

    Jie Hu, Li Shen, Samuel Albanie, et al. Gather-excite: Exploiting feature context in convolutional neural net- works. InAdvances in neural information processing systems, pages 9423–9433, 2018

    2018

  13. [13]

    Squeeze-and- excitation networks.Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and- excitation networks.Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

    2018

  14. [14]

    Adaptive mixtures of local experts.Neural Computation, 3(1):79–87, 1991

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts.Neural Computation, 3(1):79–87, 1991

    1991

  15. [15]

    Ultralytics yolov5, 2020

    Glenn Jocher. Ultralytics yolov5, 2020

    2020

  16. [16]

    Ultralytics yolo11, 2024

    Glenn Jocher and Jing Qiu. Ultralytics yolo11, 2024

    2024

  17. [17]

    Ultra- lytics yolov8, 2023

    Glenn Jocher, Ayush Chaurasia, and Jing Qiu. Ultra- lytics yolov8, 2023

    2023

  18. [18]

    Yolov13: Real-time object detection with hypergraph- enhanced adaptive visual perception.arXiv preprint arXiv:2506.17733, 2025

    Mengqi Lei, Siqi Li, Yihong Wu, and et al. Yolov13: Real-time object detection with hypergraph- enhanced adaptive visual perception.arXiv preprint arXiv:2506.17733, 2025

    Pith/arXiv arXiv 2025

  19. [19]

    Gshard: Scaling giant models with conditional computation and automatic sharding

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. Gshard: Scaling giant models with conditional computation and automatic sharding. InInternational Conference on Learning Representations, 2021

    2021

  20. [20]

    Gshard: Scaling giant models with conditional computation and automatic sharding

    Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, et al. Gshard: Scaling giant models with conditional computation and automatic sharding. InInternational Conference on Learning Representations, 2021

    2021

  21. [21]

    Base layers: Simplifying training of large, sparse models

    Mike Lewis, Shruti Bhosale, Tim Dettmers, et al. Base layers: Simplifying training of large, sparse models. International Conference on Machine Learning, pages 6265–6274, 2021

    2021

  22. [22]

    Yolov6: A single- stage object detection framework for industrial appli- cations.arXiv preprint arXiv:2209.02976, 2022

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

    Pith/arXiv arXiv 2022

  23. [23]

    Network in network

    Min Lin, Qiang Chen, and Shuicheng Yan. Network in network. InInternational Conference on Learning Representations, 2014

    2014

  24. [24]

    Lawrence Zitnick

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ´ar, and C. Lawrence Zitnick. Microsoft COCO: Com- mon objects in context. InComputer Vision – ECCV 2014, pages 740–755. Springer International Publish- ing, 2014

    2014

  25. [25]

    Feature pyramid networks for object detection.arXiv preprint arXiv:1612.03144, 2017

    Tsung-Yi Lin, Piotr Doll’ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection.arXiv preprint arXiv:1612.03144, 2017

    Pith/arXiv arXiv 2017

  26. [26]

    Swin trans- former: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, et al. Swin trans- former: Hierarchical vision transformer using shifted windows. InProceedings of the IEEE/CVF interna- tional conference on computer vision, pages 10012– 10022, 2021

    2021

  27. [27]

    Small object detection: A com- prehensive survey on challenges, techniques and real- world applications.Array, 25:100421, 2025

    Tuan Nguyen, Thanh Mai, Hai Tran, Tuan Nguyen, and Duc Nguyen. Small object detection: A com- prehensive survey on challenges, techniques and real- world applications.Array, 25:100421, 2025

    2025

  28. [28]

    Chained-tracker: Chaining paired attentive regression results for end- to-end joint multiple-object detection and tracking

    Jinlong Peng, Changan Wang, Fangbin Wan, Yang Wu, Yabiao Wang, Ying Tai, Chengjie Wang, Jilin Li, Feiyue Huang, and Yanwei Fu. Chained-tracker: Chaining paired attentive regression results for end- to-end joint multiple-object detection and tracking. In European conference on computer vision, pages 145–

  29. [29]

    From sparse to soft mixtures of experts

    Joan Puigcerver, Carlos Riquelme, Basil Mustafa, and Neil Houlsby. From sparse to soft mixtures of experts. arXiv preprint arXiv:2308.00951, 2023

    Pith/arXiv arXiv 2023

  30. [30]

    Yolov3: An incremental improvement.arXiv preprint arXiv:1804.02767, 2018

    Joseph Redmon and Ali Farhadi. Yolov3: An incremental improvement.arXiv preprint arXiv:1804.02767, 2018

    Pith/arXiv arXiv 2018

  31. [31]

    You only look once: Unified, real-time object detection.arXiv preprint arXiv:1506.02640, 2015

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection.arXiv preprint arXiv:1506.02640, 2015

    Pith/arXiv arXiv 2015

  32. [32]

    You only look once: Unified, real-time object detection.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 779–788, 2016

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection.Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 779–788, 2016

    2016

  33. [33]

    Scaling vision with sparse mixture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021

    Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, Andr’e Susano Pinto, Julian Menick, Yee Whye Xu, Jasper Snoek, Tao Yang, et al. Scaling vision with sparse mixture of experts.Advances in Neural Information Processing Systems, 34:8583–8595, 2021

    2021

  34. [34]

    Yolo advances to its genesis: a decadal and com- prehensive review of the you only look once (yolo) series.Artificial Intelligence Review, 2025

    Ranjan Sapkota, Rizwan Mishra, Meenu Yu, and Han Yu. Yolo advances to its genesis: a decadal and com- prehensive review of the you only look once (yolo) series.Artificial Intelligence Review, 2025

    2025

  35. [35]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer.arXiv preprint arXiv:1701.06538, 2017

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer.arXiv preprint arXiv:1701.06538, 2017

    Pith/arXiv arXiv 2017

  36. [36]

    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

    Pith/arXiv arXiv 2025

  37. [37]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural Information Processing Systems, 30, 2017

    2017

  38. [38]

    Yolov10: Real-time end-to-end object detection.arXiv preprint arXiv:2405.14458, 2024

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, and Guiguang Ding. Yolov10: Real-time end-to-end object detection.arXiv preprint arXiv:2405.14458, 2024

    Pith/arXiv arXiv 2024

  39. [39]

    Mamba-yolo- world: marrying yolo-world with mamba for open- vocabulary detection

    Haoxuan Wang, Qingdong He, Jinlong Peng, Hao Yang, Mingmin Chi, and Yabiao Wang. Mamba-yolo- world: marrying yolo-world with mamba for open- vocabulary detection. InICASSP 2025-2025 IEEE in- ternational conference on acoustics, speech and sig- nal processing (ICASSP), pages 1–5. IEEE, 2025

    2025

  40. [40]

    Residual mixture of experts.arXiv preprint arXiv:2204.09636, 2022

    Lemeng Wang, Soyoung Yoon, Trung Jin, James Li, Xuwang Wang, and Rose Yu Chen. Residual mixture of experts.arXiv preprint arXiv:2204.09636, 2022

    Pith/arXiv arXiv 2022

  41. [41]

    Eca- net: Efficient channel attention for deep convolutional neural networks

    Qilong Wang, Banggu Wu, Pengfei Zhu, et al. Eca- net: Efficient channel attention for deep convolutional neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 11534–11542, 2020

    2020

  42. [42]

    Cbam: Convolutional block attention module.Proceedings of the European Conference on Computer Vision (ECCV), pages 3–19, 2018

    Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon. Cbam: Convolutional block attention module.Proceedings of the European Conference on Computer Vision (ECCV), pages 3–19, 2018

    2018

  43. [43]

    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. 2023

    2023

  44. [44]

    Distance-iou loss: Faster and better learning for bounding box regression

    Zhaohui Zheng, Ping Wang, Wei Liu, Jinze Li, Rong- guang Ye, and Dongwei Ren. Distance-iou loss: Faster and better learning for bounding box regression. Proceedings of the AAAI Conference on Artificial In- telligence, 34(07):12993–13000, 2020

    2020

  45. [45]

    Mixture- of-experts with expert choice routing

    Yanqi Zhou, Tao Lei, Hanxiao Liu, et al. Mixture- of-experts with expert choice routing. InAdvances in Neural Information Processing Systems, pages 7103– 7114, 2022

    2022