Pith. sign in

REVIEW 4 major objections 5 minor 47 references

E-ConvNeXt makes ConvNeXt lightweight by splitting every stage into two streams

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 ↗

E-ConvNeXt combines CSPNet, batch-normalized ConvNeXt blocks, a stepped stem, and ESE attention to reach 78.3-81.9% ImageNet top-1 at 0.9-3.1 GFLOPs.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A plausible lightweight ConvNeXt variant, but the FLOPs arithmetic in Section 3.1.1 is wrong, so the central trade-off claim is not yet established. the 4 major comments →

arxiv 2508.20955 v1 pith:IP4UBYIK submitted 2025-08-28 cs.CV

E-ConvNeXt: A Lightweight and Efficient ConvNeXt Variant with Cross-Stage Partial Connections

Fang Wang , Huitao Li , Wenhan Chao , Zheng Zhuo , Yiran Ji , Chang Peng , Yupeng Sun This is my paper
classification cs.CV
keywords lightweight networksConvNeXtCSPNetcross-stage partial connectionsImageNet classificationchannel attentionobject detectionefficient CNNs
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.

The reading

This paper tries to make ConvNeXt usable where compute is scarce. It grafts Cross Stage Partial (CSP) connections onto ConvNeXt, splitting each stage's feature map so only part flows through the expensive blocks, and couples that with a stepped stem, batch normalization in place of layer normalization, and a lightweight channel-attention (ESE) module replacing Layer Scale. The result, E-ConvNeXt, claims network complexity reductions of up to 80% versus ConvNeXt-tiny at comparable or better accuracy for its size. The authors report ImageNet-1K top-1 accuracy of 78.3% at 0.93 GFLOPs (mini), 80.6% at 2.04 GFLOPs (tiny), and 81.9% at 3.12 GFLOPs (small), and show the backbone transfers to object detection on sonar and underwater optical images. A sympathetic reader should care because the paper is a direct recipe for turning a strong general-purpose backbone into lightweight variants without abandoning its design.

Core claim

The paper's central claim is that ConvNeXt's accuracy-efficiency deficit in small models is structural, not a scaling problem, and that integrating CSPNet's cross-stage partial routing into ConvNeXt repairs it. In E-ConvNeXt, each stage's feature map is split: one part passes through 7x7 depthwise conv plus two 1x1 convs, the other bypasses the blocks, and a 1x1 conv merges the streams; a transition hyperparameter, ch_mid = (ch_in + ch_out)/2, sets the width of the expensive stream and concentrates the FLOP reduction. Three supporting changes recover accuracy and speed: a stepped stem (2x2 conv followed by two 3x3 convs) instead of a single 4x4 patchify stem; batch normalization replacing la

What carries the argument

Cross-stage partial (CSP) routing is the load-bearing mechanism. Each CSPConvStage splits the incoming feature map into two streams; one stream goes through N ConvNeXt-style blocks (7x7 depthwise conv plus two 1x1 convs, inverted-residual structure, with BN and GELU), the other stream bypasses the blocks; a final 1x1 conv merges the streams. The transition hyperparameter ch_mid = (ch_in + ch_out)/2 is what makes the expensive stream narrower, and the bypass stream is what preserves information flow at reduced cost. The stepped stem, the BN-for-LN substitutions, and the ESE channel-attention module are supporting modifications that recover accuracy and speed lost when the network is thinned.

Load-bearing premise

The comparison table assumes that accuracy and FLOPs quoted from other papers are directly comparable to E-ConvNeXt's numbers without re-running those baseline models under a unified training recipe and a single FLOPs convention.

What would settle it

Retrain FasterNet-T2, EdgeNeXt-S, MobileViT-S, StarNet-S4, and ConvNeXt-T under the exact E-ConvNeXt training schedule and count FLOPs with one tool at 224x224; if any baseline matches or beats E-ConvNeXt at equal or lower FLOPs in the same band, the paper's accuracy-efficiency claim fails. A second check: measure wall-clock latency for E-ConvNeXt-tiny versus ConvNeXt-tiny on a representative mobile CPU; the paper predicts a 20-25% speedup for the BN block, so a substantially slower result would undercut the efficiency claim.

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

If this is right

  • E-ConvNeXt variants give a ready-made small-model family for resource-constrained deployments: 7.6M parameters at 0.93 GFLOPs, 13.2M at 2.04 GFLOPs, and 19.4M at 3.12 GFLOPs, all trained with the same 300-epoch ImageNet-1K recipe.
  • As a backbone for PP-YOLOE and YOLOv10 detectors, E-ConvNeXt-tiny raises mAP on sonar images and underwater optical images relative to the default PaNet backbone, with lower or comparable FLOPs than the larger detector configuration.
  • If the ImageNet-1K results hold under matched training conditions, E-ConvNeXt becomes a strong pure-convolution lightweight option in the 1-3 GFLOP band, making the CSP-split-plus-stepped-stem recipe a template for other heavy backbones.
  • Replacing Layer Scale with a normalized squeeze-and-excitation block suggests that channel attention can substitute for per-channel learned scaling in ConvNeXt-style blocks without destabilizing training, opening that substitution to other architectures.
  • The reported BN-for-LN swap is claimed to add no FLOPs or parameters while improving speed by about 20-25%, which would make it a nearly free efficiency upgrade for ConvNeXt-derived blocks.

Where Pith is reading between the lines

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

  • The CSP split with ch_mid creates an asymmetric information path that may behave like a learned residual gating: one testable hypothesis is that the bypass stream carries global or low-frequency context while the block stream refines details, which could be checked with feature visualizations.
  • Because the paper reports FLOPs savings but only a small FPS comparison, a practical deployment test on mobile CPUs or GPUs with measured latency would be a stronger claim than FLOPs alone; BN and 1x1-conv blocks may or may not translate to real speedups on different hardware.
  • The same recipe (CSP split + stepped stem + BN + ESE attention) could plausibly be applied to other ConvNeXt-derived or hybrid transformer backbones; the paper does not test this, but the mechanism is architecture-agnostic.
  • The 0.7% accuracy gain attributed to stage-ratio adjustment suggests that how FLOPs are distributed across stages matters as much as total FLOPs; a systematic depth-width allocation study could extend this result beyond the three variants presented.
Share X Bluesky LinkedIn Reddit HN

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

4 major / 5 minor

Summary. The paper proposes E-ConvNeXt, a lightweight ConvNeXt variant that integrates CSPNet-style cross-stage partial connections, a stepped stem, Batch Normalization in place of Layer Normalization, and an ESE channel-attention module in place of Layer Scale. Three variants (mini/tiny/small) are reported on ImageNet-1K with claimed accuracies of 78.3% at 0.93 GFLOPs, 80.6% at 2.04 GFLOPs, and 81.9% at 3.12 GFLOPs. Transfer experiments on sonar and underwater optical object detection (PP-YOLOE and YOLOv10) are reported. The central claim is a superior accuracy-efficiency frontier relative to comparable lightweight models.

Significance. If the reported accuracy-efficiency numbers are correct and reproducible, the paper offers a useful practical contribution: a simple way to adapt a well-known architecture family to resource-constrained settings, with measured ImageNet-1K results and transfer detection results. The architecture choices are plausible and the experimental protocol is conventional. However, the significance is currently conditional on resolving the FLOPs accounting and on the comparability of Table 8 baselines. The paper ships no code or configs and the GitHub link is not verifiable from the manuscript.

major comments (4)
  1. [§3.1.1] The FLOPs derivation is internally inconsistent. For the stated CSPConvNeXt block (7×7 DW on 48 channels; MLP 48→192→48 at 56×56), the terms should be roughly 56×56×49×48 ≈ 7.4M, 48×3136×192 ≈ 28.9M, and 192×3136×48 ≈ 28.9M, totaling ≈65.2M. The manuscript instead writes the second 1×1 term as '384 × 56 × 56 × 1² × 96 ≈ 29.5M'—a product of ≈115.6M—and the third term as '384 × 56 × 56 × 1² × 96 ≈ 29.5M', which is also inconsistent with the 48/192 channel configuration described in the text and Figure 2. The ConvNeXt-block total (≈257M) is also wrong (the three listed terms sum to ≈247M), so the claimed '60% reduction' does not follow from the shown arithmetic. Since Table 8’s reported GFLOPs (0.93/2.04/3.12) are central to the SOTA claim, the paper needs either a corrected derivation or—preferably—profiler-based FLOPs and fully specified configs to verify the reported numbers.
  2. [Table 8 / §4.6] The comparisons quote accuracy and FLOPs for baseline models from their original papers without a unified training or measurement protocol. Differences in training recipe (epochs, augmentation, EMA, input resolution) and FLOPs conventions (including or excluding multiply-add details, detection of efficient operators) can shift the frontier substantially. The claim that E-ConvNeXt establishes a new SOTA accuracy-FLOPs balance is therefore not yet established. At minimum the authors should disclose the exact FLOPs-measurement tool and conventions used for their own models, and for close baselines such as StarNet-S4 (78.4% at 1.07G vs. 78.3% at 0.93G), report whether the comparison is within noise. This is a load-bearing point because the paper’s main conclusion is based on this table.
  3. [§5.3.1 / Table 11 and §5.4.1 / Table 15] The object-detection comparisons mix detector scales in a way that undermines the stated conclusions. Table 11 compares YOLOv10-L with PaNet against YOLOv10-M with ConvNeXt-T and E-ConvNeXt-T; the reported FLOPs differences reflect both the backbone and the detector head, so the 'higher mAP with smaller FLOPs' claim is not attributable to the backbone. The same mixing appears in Table 15 (YOLOv10-L PaNet vs. YOLOv10-M E-ConvNeXt-T). To support the backbone-generalization claim, comparisons should hold the detector scale fixed (e.g., YOLOv10-M with PaNet vs. YOLOv10-M with E-ConvNeXt-T) or at least report paired configurations. As written, these tables support only a weaker claim that E-ConvNeXt works as a backbone, not that it outperforms PaNet at equal total FLOPs.
  4. [§3.1.3 and Table 3 vs. §3.4] There is an unresolved discrepancy in the reported FLOPs trajectory. §3.1.3 says the final CSPConvNeXt has 'the same FLOPs' as original CSPConvNeXt (1.5G per Table 3), yet the final E-ConvNeXt models are reported at 0.93G–3.12G with stage ratios and channel widths that do not match the 'input channels [65, 128, 256, 512], output channels [128, 256, 512, 1024]' configuration, since channel counts must be integers and the stem/block changes alter the totals. The paper never gives a complete configuration table (channel widths and block counts per stage) for mini/tiny/small, so the reader cannot reproduce the reported FLOPs or accuracies. This is a load-bearing documentation gap for a paper whose contribution is an efficiency frontier.
minor comments (5)
  1. [Throughout] Many typos and inconsistent naming: 'ConNeXt' vs 'ConvNeXt', 'CSPConNeXt', 'Dpethwise', 'proformence', 'Tint', 'PaNet' presumably 'PANet', 'Table 9/10 both say training data' for the test table, and 'E-ConvNeXt-Small' vs 'E-ConvNeXt-S'. Copy-editing is needed.
  2. [§2.3 and §3.3] The claim that Layer Scale is replaced by channel attention to improve accuracy needs a baseline ablation: Table 7 compares attention modules on ImageNet-100, but there is no comparison of ESE Block against the original Layer Scale (or against no scaling) at the same dataset and budget. The causal attribution in the abstract is therefore not directly supported.
  3. [§4.1 and Table 2] Training details are incomplete: the paper says 'for more detailed settings, please refer to the github repository' but the repository link is not shown to be accessible and no code/checkpoints are provided. ImageNet-100 is described as a subset but the class-selection procedure is not given.
  4. [Figure 9] The trade-off curve is difficult to read from the text-only manuscript; the figure is not included in the text. If it is included in the camera-ready, please ensure all Table 8 models are plotted with consistent FLOPs conventions.
  5. [Section 5] The sonar dataset and DUO are niche; the DUO results are labeled 'COCO underwater' in Table 15 but DUO is not a COCO subset. Please correct the label and provide the DUO test/evaluation details (e.g., whether the official metric was used).

Circularity Check

0 steps flagged

No circularity found: E-ConvNeXt's reported accuracies and FLOPs are measured or computed from explicit architecture formulas, not derived from assumptions that already contain the conclusions.

full rationale

E-ConvNeXt is an empirical architecture paper: the headline quantities (ImageNet Top-1 accuracies, FLOPs, detection mAP) are either measured outcomes or computed from standard convolution-FLOP formulas applied to a stated architecture. There is no self-definitional step, no fitted parameter renamed as a prediction, and no load-bearing self-citation; references to CSPNet, ESE, SENet, etc. are external prior work. The authors select stem, stage ratios, normalization, and attention modules using ImageNet-1K/100 experiments before reporting final numbers, which is model selection on the benchmark—a generalizability concern, not circular reasoning. The paper's own Limitations section concedes that extreme lightweight scenarios need further verification, which is a limitation, not a circular step. One non-circular correctness issue is notable: in Section 3.1.1, the FLOPs arithmetic is internally inconsistent (the third CSPConvNeXt term is written as 384×56×56×1²×96 ≈ 29.5M, but that product is ≈115.6M), so the claimed 60% reduction is not established by the shown equations; this is a burden-of-proof/correctness problem, not circularity. Likewise, Table 8 relies on externally quoted FLOPs/accuracies without re-running baselines, a comparability risk, but not a circular derivation. The central claims are self-contained against external benchmarks, and no prediction reduces by construction to its inputs.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities. It relies on hand-chosen architecture hyperparameters (ch_mid, stage ratios, stem widths) and domain assumptions about benchmark validity and training-protocol fairness. The central empirical results are measurements, so the ledger burden is moderate.

free parameters (3)
  • transition hyperparameter ch_mid = (ch_in+ch_out)/2
    Introduced in Section 3.1.2 to halve the intermediate channel count, chosen by hand to reduce FLOPs by 40%.
  • stage ratios for CSPConvStages = input [65,128,256,512], output [128,256,512,1024], blocks [3,3,9,3]
    Adjusted by hand in Section 3.1.3 to balance FLOPs and accuracy; exact values are not derived.
  • stepped stem channel widths = 3->32->64
    Chosen manually in Section 3.2.1 to combine Patchify and ResNet-vc style stems.
axioms (5)
  • standard math Convolution FLOPs formula F = Cin*Hout*Wout*K^2*Cout
    Used in Section 3.1.1 to compute block FLOPs; the arithmetic is applied inconsistently.
  • domain assumption ImageNet-1K Top-1 accuracy is a valid proxy for model quality
    The paper's central comparisons rely on ImageNet-1K accuracy as the quality metric.
  • domain assumption CSPNet's split-and-concat gradient hypothesis transfers to ConvNeXt
    Assumed in Section 3.1.1 without ablation isolating the mechanism.
  • domain assumption Batch Normalization is a drop-in replacement for Layer Normalization in ConvNeXt blocks
    The paper swaps BN for LN in blocks (Section 3.2.2) claiming no change in FLOPs or parameters; no theoretical guarantee is given.
  • domain assumption A 36-epoch training schedule is fair for YOLOv10 and PP-YOLOE with all backbones
    Section 5.1 sets 36 epochs because the original YOLOv10 usually needs hundreds; the effect on each backbone is unverified.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of E-ConvNeXt: A Lightweight and Efficient ConvNeXt Variant with Cross-Stage Partial Connections." pith.science (2026). https://pith.science/paper/IP4UBYIK

@misc{pith2026250820955,
  author       = {Pith},
  title        = {Pith review of: E-ConvNeXt: A Lightweight and Efficient ConvNeXt Variant with Cross-Stage Partial Connections},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IP4UBYIK}},
  note         = {Machine review of arXiv:2508.20955}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Many high-performance networks were not designed with lightweight application scenarios in mind from the outset, which has greatly restricted their scope of application. This paper takes ConvNeXt as the research object and significantly reduces the parameter scale and network complexity of ConvNeXt by integrating the Cross Stage Partial Connections mechanism and a series of optimized designs. The new network is named E-ConvNeXt, which can maintain high accuracy performance under different complexity configurations. The three core innovations of E-ConvNeXt are : (1) integrating the Cross Stage Partial Network (CSPNet) with ConvNeXt and adjusting the network structure, which reduces the model's network complexity by up to 80%; (2) Optimizing the Stem and Block structures to enhance the model's feature expression capability and operational efficiency; (3) Replacing Layer Scale with channel attention. Experimental validation on ImageNet classification demonstrates E-ConvNeXt's superior accuracy-efficiency balance: E-ConvNeXt-mini reaches 78.3% Top-1 accuracy at 0.9GFLOPs. E-ConvNeXt-small reaches 81.9% Top-1 accuracy at 3.1GFLOPs. Transfer learning tests on object detection tasks further confirm its generalization capability.

Figures

Figures reproduced from arXiv: 2508.20955 by Chang Peng, Fang Wang, Huitao Li, Wenhan Chao, Yiran Ji, Yupeng Sun, Zheng Zhuo.

Figure 1
Figure 1. Figure 1: E-ConvNeXt Evolution Figure shows the steps from ConvNeXt to E-ConvNeXt. The bar chart represents the Top￾1 accuracy on ImageNet-1K, and the line chart shows the changes in model FLOPs. 2. Optimization of the ConvNeXt structure: 1) A stepped stem is proposed to alleviate the problem of infor￾mation loss caused by continuous downsampling. 2) replacing Layer Normalization[27] (LN) with Batch Normalization[28… view at source ↗
Figure 3
Figure 3. Figure 3: (A) is Original CSPConvNeXt CSP Stage. (B) is 𝑐ℎmid CSPConvNeXt CSP Stage complexity decreases, the model accuracy exhibits a cor￾responding decline. To achieve a balance between FLOPs and accuracy, we increase the FLOPs of the 𝑐ℎ𝑚𝑖𝑑 CSP￾ConvNeXt to the same level as the Original CSPConvNeXt by adjusting the stage ratio. The adjusted stage ratio is as follows: input channels are [65, 128, 256, 512], output… view at source ↗
Figure 2
Figure 2. Figure 2: CSPConvNets demonstration figure shows how CSPNet transforms ResNet into CSPResNet, and how we transform ConvNeXt into CSPConvNeXt. replace the Split operation, which can better split the feature map. The feature map channel transformation process of the original CSPNet is 𝑐ℎin down￾sampling ←←←←←←←←←←←←←←←←←←←←←←←←←←←←←→ 𝑐ℎout Split ←←←←←←←←←←←←←←←→ 2 × ( 𝑐ℎout 2 ) concat ←←←←←←←←←←←←←←←←←←←←←←→ 𝑐ℎout To … view at source ↗
Figure 4
Figure 4. Figure 4: After adjusting the Stage ratio, CSPConvNeXt consists of one Stem and four CSPConvStages. Each Stage includes a downsampling layer and multiple CSPConvNeXt Blocks. The output of each Stage is twice the size of its input. The diagram shows the channel configuration of CSPConvNeXt-tiny. of spatial features. The final stem structure is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison diagram of Block structures: The Con￾vNeXt Block uses fully connected layers to form the MLP layer, with Layer Normalization (LN) for normalization. The CSPConvNeXt Block employs 1x1 convolutions to construct the MLP layer, and adopts Batch Normalization (BN) for normalization. Conv 1x1 DConv 7x7 Conv 1x1 SE Block1 SE Block BatchNorm GELU BatchNorm Conv 1x1 DConv 7x7 Conv 1x1 SE Block2 BatchNorm… view at source ↗
Figure 5
Figure 5. Figure 5: a is the Stem of ConvNeXt; b is the Stem of CSP￾ConvNeXt after Stage ratio adjustment and the downsampling layer in the first CSPStage; c is the Stem layer of 𝑅𝑒𝑠𝑁𝑒𝑡vc; d is the Step Stem. The Step Stem is formed by integrating the characteristics of the Stem layers of ConvNeXt and 𝑅𝑒𝑠𝑁𝑒𝑡vc. by splitting the feature map into two parts.ne feature map passes through the blocks and the other does not, and the… view at source ↗
Figure 7
Figure 7. Figure 7: SE Block1 adopts the integration method of SENet, with the number of channels in the SE Block being c. SE Block2 follows the integration method of EfficientNet, where the number of channels in the SE Block is 4c. 4. Experiments This paper used two benchmark datasets: ImageNet1K[10] and ImageNet-1001 . • ImageNet-1K: It is widely used in image classification tasks. It consists of 1,000 classes and contains … view at source ↗
Figure 8
Figure 8. Figure 8: The overall architecture of E-ConvNeXt, illustrating the hierarchical stages, including the Stepped Stem, CSPConvStages, and the integration of key components such as ESE Block, Batch Normalization, and depthwise convolutions [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Rade-off curves figure. E-ConvNeXt achieves the highest efficiency in balancing accuracy and FLOPs. fully demonstrate the application prospects of E-ConvNeXt in lightweight scenarios. 5. E-ConvNeXt on downstream tasks We conducted extensive experiments to verify the per￾formance of E-ConvNeXt in object detection. The exper￾iments were carried out on the sonar image dataset and the DUO dataset[45]. In the e… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

47 extracted references · 31 canonical work pages · 1 internal anchor

  1. [1]

    Back- propagation applied to handwritten zip code recognition

    Yann LeCun, Bernhard Boser, John S Denker, Donnie Henderson, Richard E Howard, Wayne Hubbard, and Lawrence D Jackel. Back- propagation applied to handwritten zip code recognition. Neural computation, 1(4):541–551, 1989

    work page 1989

  2. [2]

    Imagenet classification with deep convolutional neural networks.Advances in neural information processing systems, 25, 2012

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks.Advances in neural information processing systems, 25, 2012

    2012

  3. [3]

    Deep residual learning for image recognition

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

    2016

  4. [4]

    Efficientnet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. InInternational conference on machine learning, pages 6105–6114. PMLR, 2019

    2019

  5. [5]

    At- tention is all you need.Advances in neural information processing systems, 30, 2017

    AshishVaswani,NoamShazeer,NikiParmar,JakobUszkoreit,Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. At- tention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  6. [6]

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

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weis- senborn, 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. arXiv preprint arXiv:2010.11929, 2020

    Pith/arXiv arXiv 2010

  7. [7]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11976–11986, 2022

    2022

  8. [8]

    F Wang et al.:Preprint submitted to Elsevier Page 11 of 13 E-ConvNeXt Mobilenets:Efficientconvolutionalneuralnetworksformobilevision applications

    Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, WeijunWang,TobiasWeyand,MarcoAndreetto,andHartwigAdam. F Wang et al.:Preprint submitted to Elsevier Page 11 of 13 E-ConvNeXt Mobilenets:Efficientconvolutionalneuralnetworksformobilevision applications. arXiv preprint arXiv:1704.04861, 2017

    Pith/arXiv arXiv 2017

  9. [9]

    Swintransformer:Hierarchicalvision transformerusingshiftedwindows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, StephenLin,andBainingGuo. Swintransformer:Hierarchicalvision transformerusingshiftedwindows. In ProceedingsoftheIEEE/CVF international conference on computer vision, pages 10012–10022, 2021

    work page 2021

  10. [10]

    Imagenet large scale visual recognition challenge.Internationaljournalofcomputervision ,115(3):211–252, 2015

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh,SeanMa,ZhihengHuang,AndrejKarpathy,AdityaKhosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge.Internationaljournalofcomputervision ,115(3):211–252, 2015

    work page 2015

  11. [11]

    Semantic understanding of scenes through the ade20k dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Tete Xiao, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Semantic understanding of scenes through the ade20k dataset. International Journal of Computer Vision, 127(3):302–321, 2019

    work page 2019

  12. [12]

    Mi- crosoftcoco:Commonobjectsincontext

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Mi- crosoftcoco:Commonobjectsincontext. In Europeanconferenceon computer vision, pages 740–755. Springer, 2014

    work page 2014

  13. [13]

    Mobilenetv2: Inverted residuals and linear bottlenecks

    MarkSandler,AndrewHoward,MenglongZhu,AndreyZhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4510–4520, 2018

    work page 2018

  14. [14]

    Searching for mobilenetv3

    Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, et al. Searching for mobilenetv3. InProceedings of the IEEE/CVF international conference on computer vision, pages 1314–1324, 2019

    2019

  15. [15]

    Run, don’t walk: chasing higher flops for faster neural networks

    Jierun Chen, Shiu-hong Kao, Hao He, Weipeng Zhuo, Song Wen, Chul-Ho Lee, and S-H Gary Chan. Run, don’t walk: chasing higher flops for faster neural networks. InProceedings of the IEEE/CVF conferenceoncomputervisionandpatternrecognition ,pages12021– 12031, 2023

    work page 2023

  16. [16]

    Cspnet: A new backbone that can enhance learning capability of cnn

    Chien-Yao Wang, Hong-Yuan Mark Liao, Yueh-Hua Wu, Ping-Yang Chen, Jun-Wei Hsieh, and I-Hau Yeh. Cspnet: A new backbone that can enhance learning capability of cnn. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 390–391, 2020

    2020

  17. [17]

    Shuf- flenet:Anextremelyefficientconvolutionalneuralnetworkformobile devices, 2017

    Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, and Jian Sun. Shuf- flenet:Anextremelyefficientconvolutionalneuralnetworkformobile devices, 2017

    work page 2017

  18. [18]

    Shuf- flenet:Anextremelyefficientconvolutionalneuralnetworkformobile devices

    Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, and Jian Sun. Shuf- flenet:Anextremelyefficientconvolutionalneuralnetworkformobile devices. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 6848–6856, 2018

    work page 2018

  19. [19]

    In International conference on machine learning, pages 10096–10106

    MingxingTanandQuocLe.Efficientnetv2:Smallermodelsandfaster training. In International conference on machine learning, pages 10096–10106. PMLR, 2021

    work page 2021

  20. [20]

    Densely connected convolutional networks

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. In Pro- ceedings of the IEEE conference on computer vision and pattern recognition, pages 4700–4708, 2017

    work page 2017

  21. [21]

    Aggregatedresidualtransformationsfordeepneuralnetworks

    Saining Xie, Ross Girshick, Piotr Dollár, Zhuowen Tu, and Kaiming He. Aggregatedresidualtransformationsfordeepneuralnetworks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1492–1500, 2017

    work page 2017

  22. [22]

    Squeeze-and-excitationnetworks

    JieHu,LiShen,andGangSun. Squeeze-and-excitationnetworks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018

    work page 2018

  23. [23]

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

    Qilong Wang, Banggu Wu, Pengfei Zhu, Peihua Li, Wangmeng Zuo, and Qinghua Hu. Eca-net: Efficient channel attention for deep convolutional neural networks. In Proceedings of the IEEE/CVF conferenceoncomputervisionandpatternrecognition ,pages11534– 11542, 2020

    work page 2020

  24. [24]

    Global second-order pooling convolutional networks

    Zilin Gao, Jiangtao Xie, Qilong Wang, and Peihua Li. Global second-order pooling convolutional networks. InProceedings of the IEEE/CVF Conference on computer vision and pattern recognition, pages 3024–3033, 2019

    work page 2019

  25. [25]

    Srm: A style- based recalibration module for convolutional neural networks

    HyunJae Lee, Hyo-Eun Kim, and Hyeonseob Nam. Srm: A style- based recalibration module for convolutional neural networks. In Proceedings of the IEEE/CVF International conference on computer vision, pages 1854–1862, 2019

    work page 2019

  26. [26]

    Fcanet: Frequency channel attention networks

    Zequn Qin, Pengyi Zhang, Fei Wu, and Xi Li. Fcanet: Frequency channel attention networks. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 783–792, 2021

    work page 2021

  27. [27]

    Beit: Bert pre- training of image transformers

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. Beit: Bert pre- training of image transformers. arXiv preprint arXiv:2106.08254, 2021

    Pith/arXiv arXiv 2021

  28. [28]

    Batch renormalization: Towards reducing minibatch dependence in batch-normalized models.Advances in neural infor- mation processing systems, 30, 2017

    Sergey Ioffe. Batch renormalization: Towards reducing minibatch dependence in batch-normalized models.Advances in neural infor- mation processing systems, 30, 2017

    work page 2017

  29. [29]

    Centermask: Real-time anchor- free instance segmentation

    Youngwan Lee and Jongyoul Park. Centermask: Real-time anchor- free instance segmentation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 13906– 13915, 2020

    work page 2020

  30. [30]

    Original approach for the localisation of objects in images.IEE Proceedings- Vision, Image and Signal Processing, 141(4):245–250, 1994

    Régis Vaillant, Christophe Monrocq, and Yann Le Cun. Original approach for the localisation of objects in images.IEE Proceedings- Vision, Image and Signal Processing, 141(4):245–250, 1994

    work page 1994

  31. [31]

    Decoupled weight decay regular- ization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regular- ization. arXiv preprint arXiv:1711.05101, 2017

    Pith/arXiv arXiv 2017

  32. [32]

    Randaugment:Practicalautomateddataaugmentationwithareduced search space

    Ekin D Cubuk, Barret Zoph, Jonathon Shlens, and Quoc V Le. Randaugment:Practicalautomateddataaugmentationwithareduced search space. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, pages 702–703, 2020

    work page 2020

  33. [33]

    mixup: Beyond empirical risk minimization

    HongyiZhang,MoustaphaCisse,YannNDauphin,andDavidLopez- Paz. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412, 2017

    Pith/arXiv arXiv 2017

  34. [34]

    Cutmix: Regularization strategy to train strong classifiers with localizable features

    Sangdoo Yun, Dongyoon Han, Seong Joon Oh, Sanghyuk Chun, Junsuk Choe, and Youngjoon Yoo. Cutmix: Regularization strategy to train strong classifiers with localizable features. InProceedings of the IEEE/CVF international conference on computer vision, pages 6023–6032, 2019

    2019

  35. [35]

    Random erasing data augmentation

    Zhun Zhong, Liang Zheng, Guoliang Kang, Shaozi Li, and Yi Yang. Random erasing data augmentation. In Proceedings of the AAAI conferenceonartificialintelligence ,volume34,pages13001–13008, 2020

    work page 2020

  36. [36]

    Rethinking the inception architecture for computer vision

    ChristianSzegedy,VincentVanhoucke,SergeyIoffe,JonShlens,and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016

    work page 2016

  37. [37]

    Ghostnet: More features from cheap operations

    Kai Han, Yunhe Wang, Qi Tian, Jianyuan Guo, Chunjing Xu, and Chang Xu. Ghostnet: More features from cheap operations. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1580–1589, 2020

    work page 2020

  38. [38]

    Shuf- flenetv2:Practicalguidelinesforefficientcnnarchitecturedesign

    Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, and Jian Sun. Shuf- flenetv2:Practicalguidelinesforefficientcnnarchitecturedesign. In ProceedingsoftheEuropeanconferenceoncomputervision(ECCV) , pages 116–131, 2018

    work page 2018

  39. [39]

    Mobilevit: light-weight, general-purpose, and mobile-friendly vision transformer

    Sachin Mehta and Mohammad Rastegari. Mobilevit: light-weight, general-purpose, and mobile-friendly vision transformer. arXiv preprint arXiv:2110.02178, 2021

    Pith/arXiv arXiv 2021

  40. [40]

    Edgenext: efficiently amalgamated cnn-transformer archi- tecture for mobile vision applications

    MuhammadMaaz,AbdelrahmanShaker,HishamCholakkal,Salman Khan, Syed Waqas Zamir, Rao Muhammad Anwer, and Fahad Shah- baz Khan. Edgenext: efficiently amalgamated cnn-transformer archi- tecture for mobile vision applications. InEuropean conference on computer vision, pages 3–20. Springer, 2022

    work page 2022

  41. [41]

    Rewrite the stars

    Xu Ma, Xiyang Dai, Yue Bai, Yizhou Wang, and Yun Fu. Rewrite the stars. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5694–5703, 2024

    work page 2024

  42. [42]

    Cyclemlp: A mlp-likearchitecturefordenseprediction.arxiv2021

    S Chen, E Xie, C Ge, R Chen, D Liang, and P Luo. Cyclemlp: A mlp-likearchitecturefordenseprediction.arxiv2021. arXivpreprint arXiv:2107.10224

    Pith/arXiv arXiv

  43. [43]

    Metaformer is actually what F Wang et al.:Preprint submitted to Elsevier Page 12 of 13 E-ConvNeXt you need for vision

    Weihao Yu, Mi Luo, Pan Zhou, Chenyang Si, Yichen Zhou, Xinchao Wang, Jiashi Feng, and Shuicheng Yan. Metaformer is actually what F Wang et al.:Preprint submitted to Elsevier Page 12 of 13 E-ConvNeXt you need for vision. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10819–10829, 2022

    work page 2022

  44. [44]

    Pyramid vision transformer: A versatile backbone for dense prediction without con- volutions

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without con- volutions. InProceedings of the IEEE/CVF international conference on computer vision, pages 568–578, 2021

    work page 2021

  45. [45]

    A Dataset And Benchmark Of Underwater Object Detection For Robot Picking

    Chongwei Liu, Haojie Li, Shuchang Wang, Ming Zhu, Dong Wang, Xin Fan, and Zhihui Wang. A Dataset And Benchmark Of Un- derwater Object Detection For Robot Picking.arXiv e-prints, page arXiv:2106.05681, June 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  46. [46]

    arXivpreprint arXiv:2203.16250, 2022

    ShangliangXu,XinxinWang,WenyuLv,QinyaoChang,ChengCui, Kaipeng Deng, Guanzhong Wang, Qingqing Dang, Shengyu Wei, YuningDu,etal.Pp-yoloe:Anevolvedversionofyolo. arXivpreprint arXiv:2203.16250, 2022

    Pith/arXiv arXiv 2022

  47. [47]

    Yolov10: Real-time end-to-end object detection.Advances in Neural Information Processing Systems, 37:107984–108011, 2024

    Ao Wang, Hui Chen, Lihao Liu, Kai Chen, Zijia Lin, Jungong Han, et al. Yolov10: Real-time end-to-end object detection.Advances in Neural Information Processing Systems, 37:107984–108011, 2024. F Wang et al.:Preprint submitted to Elsevier Page 13 of 13

    work page 2024

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.