Pith. sign in

REVIEW 4 major objections 7 minor 44 references

VisionGRU: A Linear-Complexity RNN Model for Efficient Image Analysis

T0 review · 4 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read VisionGRU claims that a linear-complexity recurrent backbone using minGRU units and bidirectional raster scanning matches or beats vision transformers on image classification and semantic segmentation while cutting compute and memory…

desk verdict A credible RNN-vision backbone whose real numbers are buried under overstated abstract and intro claims; worth reviewing but needs honest comparisons, ablations, and code. read the letter →

arxiv 2412.18178 v2 pith:GPDT4DMS submitted 2024-12-24 cs.CV

Shicheng Yin , Kaixuan Yin , Weixing Chen , Enbo Huang , Yang Liu This is my paper · ORCID
classification cs.CV
keywords VisionGRUminrecurrentneuralnetworkimageclassificationsemanticsegmentationlinearcomplexitybidirectionalscanningbackbone
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 aims to show that a recurrent neural network redesigned for 2D image data can match or beat vision transformers on classification and segmentation while using far less compute and memory. The proposed architecture, VisionGRU, uses a simplified gated recurrent unit (minGRU) to process image features in linear time, with a bidirectional raster-scanning module intended to capture global context. On ImageNet-1K, VisionGRU-Ti reaches 82.0% top-1 accuracy versus 72.2% for DeiT-Ti, and VisionGRU-B reaches 83.1% versus 81.8% for DeiT-B; on ADE20K segmentation it reports 44.7% mIoU versus 43.8% for a DeiT-based baseline. If these results hold, RNN-based backbones become a viable alternative to attention models, especially at high input resolutions where self-attention's quadratic cost becomes prohibitive.

What carries the argument

The load-bearing component is the 2DGRU module, built from minGRU units. minGRU is a simplified Gated Recurrent Unit that removes the reset gate and the dependence of the update gate and candidate state on previous hidden states, so the recurrence can be computed in parallel using a prefix-scan algorithm without backpropagation through time. In 2DGRU, a feature map is flattened along a raster scan and processed in both forward and reverse directions by two minGRU passes, then the outputs are aggregated position-wise, followed by a residual connection and a feed-forward network. The claimed effect is that every spatial location integrates information from all preceding and succeeding locations in the scan, which the paper argues provides long-range 2D context. Hierarchical downsampling layers between 2DGRU stages progressively shrink spatial dimensions and expand channels, mimicking the multi-scale structure of CNN and Swin backbones while keeping total compute linear in image size.

What would settle it

Run VisionGRU on ADE20K with only a unidirectional raster scan or with a different scan order such as column-major or a Hilbert curve; if the mIoU does not drop appreciably when bidirectionality or raster order is removed, then the bidirectional global-context mechanism is not the source of the reported segmentation gains.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a hierarchical recurrent backbone built from minGRU units can outperform vision transformers on both image classification and semantic segmentation without sacrificing efficiency. The key reported numbers are 82.0% top-1 accuracy for VisionGRU-Ti (30M params) against 72.2% for DeiT-Ti (6M) and 83.1% for VisionGRU-B (86M) against 81.8% for DeiT-B (86M), plus 44.7% mIoU on ADE20K with a UperNet head. The paper attributes these gains to the 2DGRU module, which scans feature maps in forward and reverse raster order with bidirectional minGRU layers, letting each spatial position accumulate context from the whole image. Combined with downsampling stages that reduce sequence length while widening channels, the model retains multi-scale features with computational complexity that grows linearly with image area, a direct contrast with self-attention's quadratic scaling. The paper frames this as evidence that RNNs, when simplified for parallel training and adapted to 2D structure, can close the gap with attention-based architectures that had displaced them.

Load-bearing premise

The central assumption is that flattening a 2D feature map into two flat raster scans (forward and reverse) gives each pixel enough global 2D context for the claimed accuracy gains; no experiment varies the scan order or compares against a spatially-aware alternative.

Editorial extensions

If this is right

  • If VisionGRU's results replicate, recurrent backbones become a practical choice for high-resolution vision tasks where self-attention's quadratic cost is a bottleneck.
  • The 2DGRU's bidirectional scanning could be transferred to other dense prediction tasks, such as detection or depth estimation, where global context matters.
  • The minGRU-based design implies training can be parallelized without BPTT, so scaling to larger images does not automatically increase memory as steeply as attention models.
  • The reported efficiency gains (e.g., 151.9 GFLOPs at 1248x1248 vs 432.3 for DeiT-S) suggest mobile or real-time deployments could benefit from replacing transformer backbones with this RNN.
  • The architecture's success would validate the broader direction of recurrence- and state-space-based vision models that aim to bridge the performance gap with attention.

Reading between the lines

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

  • The paper does not test whether the bidirectional raster scan is essential; a controlled comparison with unidirectional scanning, column-major scanning, or a serpentine order would isolate whether the claimed global-context benefit comes from bidirectionality or simply from scan depth.
  • The reported FLOPs advantage might not translate linearly to wall-clock speed on modern hardware, because repeated sequential dependencies and memory-access patterns in raster scans could introduce overhead; benchmarking throughput and latency at multiple resolutions would clarify the practical gain.
  • If the scan-order assumption holds, extending VisionGRU to video by scanning the spatial-temporal volume could give a linear-complexity alternative to 3D attention, though the paper does not address this.
  • The model's parameter efficiency relative to DeiT-Ti is not apples-to-apples (30M vs 6M), so a fairer comparison controlling for parameter count at the same budget would strengthen the claim that RNNs close the gap with attention.
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

4 major / 7 minor

Summary. VisionGRU proposes a hierarchical recurrent backbone for image classification and semantic segmentation. It replaces self-attention with minGRU units arranged in a 2DGRU module that scans the 2D feature map in forward and reverse raster order, with hierarchical downsampling to obtain multi-scale features. Experiments on ImageNet-1K report 82.0% top-1 accuracy for VisionGRU-Ti and 83.1% for VisionGRU-B, and on ADE20K the UperNet+VisionGRU-Ti combination reaches 44.7% mIoU. The paper claims linear computational complexity, reduced memory, and higher efficiency than DeiT models, especially at high resolution.

Significance. If the empirical claims were reproducible, the paper would make a useful contribution: it would show that a linear-complexity recurrent backbone, built from minGRU and parallel prefix-scan training, can match or exceed ViTs on image classification and dense prediction. The design choice of bidirectional raster scans, inherited from SSM-based vision models, is clearly stated, and the paper explicitly identifies the 2DGRU module as the core novelty. However, the current evidence does not support the strongest headline claims: the key comparisons in the abstract and introduction are not parameter-matched, single runs are reported without variance, and the FLOPs figure contradicts the claim of lower cost at all resolutions. With corrected reporting and additional matched-budget experiments, the contribution could become solid, but as presented the central quantitative claims need substantial revision.

major comments (4)
  1. [Abstract, Section I, Table I, Fig. 1] The headline claim that 'with the same number of parameters, VisionGRU achieves 2.2% higher accuracy in classification and 0.9% higher mIoU in segmentation while being 184% more computationally efficient' is not supported by the paper's own data. In Table I, the only parameter-matched classification comparison is VisionGRU-B (86M, 83.1%) versus DeiT-B (86M, 81.8%), which is a gain of 1.3 points, not 2.2. The 2.2-point gain (82.0 vs 79.8) compares VisionGRU-Ti at 30M parameters with DeiT-S at 22M parameters, a 36% parameter advantage. Fig. 1 repeats this mismatched pairing, and at the smallest resolution shown (512) VisionGRU-Ti is 25.6 GFLOPs versus DeiT-S's 24.6 GFLOPs, contradicting the caption's 'at all resolutions'. The 184% efficiency figure (432.3/151.9 at 1248) is a single high-resolution point. The abstract and introduction should either be corrected to report only the matched-budget comparisons, or the authors should add properly matched experiments (same parameter count and same resolution) that justify the stated numbers.
  2. [Tables I and II, Section IV] No variance or statistical reliability information is reported for any experimental result. The classification differences that drive the paper's narrative are 1.3 and 2.2 accuracy points, and the segmentation gain is 0.9 mIoU; these margins may be within run-to-run variability for ImageNet and ADE20K training. The authors should report results over multiple seeds (e.g., mean and standard deviation, or individual runs) and state whether the hyperparameters were reused from DeiT for all models or tuned for VisionGRU. Without this information, the central empirical comparison cannot be evaluated.
  3. [Section III.C, Section IV.B] The claim that the 2DGRU module 'captures both local and global context' and 'improves long-range dependency modeling' is load-bearing for the reported segmentation gains, but it is never validated by ablation. The paper assumes that two flat raster scans (forward and reverse) supply sufficient 2D global context, yet no experiment varies the scan order (e.g., row-major vs column-major), the number of scanning directions, or compares against a 2D-aware alternative. Since the mIoU improvement over DeiT is small (0.9 points) and the architecture's advantage over windowed attention is argued to come exactly from this global scanning, an ablation isolating the scanning strategy is necessary to support the design claim.
  4. [Section IV.C, Fig. 1] The efficiency claims are not substantiated. Fig. 1 reports FLOPs for a mismatched pair (VisionGRU-Ti vs DeiT-S) and uses a single point to claim a 184% improvement; it also shows that at 512 resolution VisionGRU-Ti is not cheaper. The paper does not describe how FLOPs were computed, and it reports no wall-clock training or inference time, no GPU memory measurements, and no throughput figures, despite the abstract's claims about 'significantly reducing memory usage and computational costs'. The authors should either provide a clear FLOPs formula for their architecture, report measured runtime and memory, and restrict efficiency statements to operating points where the comparison actually holds.
minor comments (7)
  1. [Section I] The introduction contains two nearly identical consecutive sentences describing the bidirectional scanning mechanism and its benefit for semantic segmentation; one should be removed.
  2. [References] Reference [26] appears to be misattributed: the text cites Feng et al. for minGRU and 'Were RNNs all we needed?', but the reference entry lists different authors and a different title. Please correct the citation.
  3. [References] References [19] and [20] both list papers titled 'Vision Mamba', which is confusing; please verify the entries and disambiguate them (e.g., with arXiv identifiers).
  4. [Section III.C] The formulas in Eqs. (10) and (11) have indexing that is hard to follow: for the forward path, y→,m is defined for m ≤ L_{i,j} but includes a product from k=m+1 to L_{i,j}, and the backward path uses a product from k=L_{i,j} to m−1. Please clarify the index ranges and state whether the sums in Eq. (9) run over all scan positions m.
  5. [Algorithm 1, Section III.C] The pseudocode uses a variable 'Paths' without defining it; please make explicit that the set is {forward raster, reverse raster} and state how the aggregation is performed.
  6. [Section III.D] The text gives block distributions [2,2,15,2] and [2,2,8,2] for Base and Tiny, respectively, which do sum to 21 and 14, but it omits the hidden dimensions and FFN expansion factors for each stage. These details are needed to reproduce the parameter counts in Table I.
  7. [Section IV.A] The training setup is said to use 'AdamW with a momentum of 0.9'; AdamW uses betas rather than momentum, so please correct the terminology or provide the exact beta values.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: VisionGRU's central claims rest on external baselines and the externally defined minGRU equations; the only self-referential material is a non-load-bearing future-work citation block.

full rationale

The paper's derivation chain is self-contained in the relevant sense. The 2DGRU module is built from minGRU, whose equations (6)-(8) and linear-complexity/parallel-training properties are taken from external prior work (Feng et al. / 'Were RNNs all we needed?'), not derived from VisionGRU's own outputs; this is inheritance, not circularity. Equations (9)-(11) algebraically unroll the minGRU recurrence in two raster directions, so the 'each output depends solely on the input features' claim is a direct expansion of the adopted recurrence rather than a prediction that reduces to a fitted parameter. The ImageNet and ADE20K comparisons are against external baselines (DeiT, Swin, Vim), and no fitted parameter is renamed as a prediction. The headline '+2.2% / 184%' claims are not well supported by the paper's own tables, but that is a reporting/evidence inconsistency, not a circularity: the reported numbers are experiments, not quantities forced by construction. The self-citations in the paper appear in general-introduction citations and in the future-directions sentence of Section IV-C ('open new directions for more works [34]-[44]'); they do not support the architecture, the equations, or the empirical comparisons, so they are not load-bearing. Under the hard rules, a non-load-bearing self-citation cluster warrants at most a 2, and no circular reduction is exhibited in the derivation chain.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The paper is an empirical architecture proposal. It introduces no new theory; it transfers minGRU, a parallelizable RNN cell, to vision by flattening feature maps into scan sequences. The central claims rest on hand-chosen stage depths, unreported hidden dimensions, standard training recipes, and the assumption that raster scans capture global context. There are no invented physical entities; the 2DGRU block is a new architectural construct supported only by the reported benchmarks.

free parameters (5)
  • Stage block counts = Base [2,2,15,2], Tiny [2,2,8,2]
    Chosen by hand; no ablation; directly sets model depth and FLOPs.
  • Hidden dimension and FFN expansion = Not reported
    Central capacity and FLOPs drivers are omitted, so reported efficiency cannot be independently reconstructed.
  • Downsampling schedule = Three stride-2 conv layers at stage boundaries
    Defines the multi-scale hierarchy and sequence-length reduction; not derived.
  • Training hyperparameters = lr 1e-3*batch/1024, EMA 0.9999, wd 0.05, 300 epochs
    Adopted from DeiT; these choices affect final accuracy but are not ablated.
  • Learned weights of stem, minGRU, conv, FFN = 30M (Ti), 86M (B)
    All model parameters are fitted to ImageNet-1K; the accuracy claims depend on them.
assumptions (5)
  • standard math minGRU recurrence supports parallel prefix-scan training with linear time complexity
    Inherited from Feng et al. [26]; used in Section III.A.3 and III.C to justify efficiency claims.
  • domain assumption Raster-order forward and reverse scans over the flattened feature map provide sufficient global context
    Stated in Section III.B and Algorithm 1; no ablation tests other scan orders or 2D-aware recurrences.
  • domain assumption UperNet is a neutral decoder for comparing backbones
    Section IV.B uses UperNet for all backbones; assumes no backbone-decoder interaction.
  • domain assumption ImageNet-1K and ADE20K labels and splits are standard and correct
    Standard benchmark assumption; not independently verified.
  • ad hoc to paper LN + depthwise conv + minGRU + FFN residual block is a sufficient design
    Design chosen without component analysis or ablations.
invented entities (1)
  • 2DGRU module
    purpose: Bidirectional minGRU scanning on 2D feature maps to capture local and global context
    Central new component; only support is the paper's own reported benchmarks; no independent validation or code.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VisionGRU: A Linear-Complexity RNN Model for Efficient Image Analysis." pith.science (2026). https://pith.science/paper/GPDT4DMS

@misc{pith2026241218178,
  author       = {Pith},
  title        = {Pith review of: VisionGRU: A Linear-Complexity RNN Model for Efficient Image Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GPDT4DMS}},
  note         = {Machine review of arXiv:2412.18178}
}
read the original abstract

Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) are two dominant models for image analysis. While CNNs excel at extracting multi-scale features and ViTs effectively capture global dependencies, both suffer from high computational costs, particularly when processing high-resolution images. Recently, state-space models (SSMs) and recurrent neural networks (RNNs) have attracted attention due to their efficiency. However, their performance in image classification tasks remains limited. To address these challenges, this paper introduces VisionGRU, a novel RNN-based architecture designed for efficient image classification. VisionGRU leverages a simplified Gated Recurrent Unit (minGRU) to process large-scale image features with linear complexity. It divides images into smaller patches and progressively reduces the sequence length while increasing the channel depth, thus facilitating multi-scale feature extraction. A hierarchical 2DGRU module with bidirectional scanning captures both local and global contexts, improving long-range dependency modeling, particularly for tasks like semantic segmentation. Experimental results on the ImageNet and ADE20K datasets demonstrate that VisionGRU outperforms ViTs, significantly reducing memory usage and computational costs, especially for high-resolution images. These findings underscore the potential of RNN-based approaches for developing efficient and scalable computer vision solutions. Codes will be available at https://github.com/YangLiu9208/VisionGRU.

Figures

Figures reproduced from arXiv: 2412.18178 by the authors.

Figure 1
Figure 1. VisionGRU-Ti achieves higher classification accuracy [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the VisionGRU model. It integrates the computational strengths of RNNs and CNNs using a hierarchical [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The semantic segmentation example from the ADE20K validation set. The left image shows the segmentation result of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 29 canonical work pages

  1. [1]

    Deep learning,

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  2. [2]

    Cross-modal causal relational reasoning for event-level visual question answering,

    Yang Liu, Guanbin Li, and Liang Lin, “Cross-modal causal relational reasoning for event-level visual question answering,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 10, pp. 11624– 11641, 2023

    work page 2023

  3. [3]

    Visual causal scene refinement for video question answering,

    Yushen Wei, Yang Liu, Hong Yan, Guanbin Li, and Liang Lin, “Visual causal scene refinement for video question answering,” in Proceedings of the 31st ACM International Conference on Multimedia , 2023, pp. 377–386

    2023

  4. [4]

    Imagenet classification with deep convolutional neural networks,

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton, “Imagenet classification with deep convolutional neural networks,” Communications of the ACM , vol. 60, no. 6, pp. 84–90, 2017

    work page 2017

  5. [5]

    An image is worth 16x16 words: Transform- ers for image recognition at scale,

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weis- senborn, and Neil Houlsby, “An image is worth 16x16 words: Transform- ers for image recognition at scale,” arXiv preprint arXiv:2010.11929 , 2020

    arXiv 2010

  6. [6]

    Transformers in vision: A survey,

    Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, and Mubarak Shah, “Transformers in vision: A survey,” ACM Computing Surveys (CSUR) , vol. 54, no. 10s, pp. 1–41, 2022

    work page 2022

  7. [7]

    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,” in Advances in neural information processing systems , 2017, vol. 30

    work page 2017

  8. [8]

    MetaLA: Unified Optimal Linear Approximation to Softmax Attention Map

    Yuhong Chou, Man Yao, Kexin Wang, Yuqi Pan, Ruijie Zhu, Yiran Zhong, Yu Qiao, Jibin Wu, Bo Xu, and Guoqi Li, “Metala: Unified optimal linear approximation to softmax attention map,” arXiv preprint arXiv:2411.10741, 2024

    work page Pith review arXiv 2024

Show all 44 references
  1. [9]

    Mamba: A scalable state-space model for sequence modeling,

    Albert Gu, Marcus Lee, Aditya Grover, Thomas L Paine, and Jascha Sohl-Dickstein, “Mamba: A scalable state-space model for sequence modeling,” Advances in Neural Information Processing Systems , 2022

    2022

  2. [10]

    Learning phrase representations using rnn encoder-decoder for statistical machine translation,

    Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio, “Learning phrase representations using rnn encoder-decoder for statistical machine translation,” arXiv preprint arXiv:1406.1078 , 2014

    2014 arXiv

  3. [11]

    Were rnns all we needed?,

    Leo Feng, Frederick Tung, Mohamed Osama Ahmed, Yoshua Bengio, and Hossein Hajimirsadeghi, “Were rnns all we needed?,” 2024

    2024

  4. [12]

    Very deep convolutional net- works for large-scale image recognition,

    Karen Simonyan and Andrew Zisserman, “Very deep convolutional net- works for large-scale image recognition,” arXiv preprint arXiv:1409.1556, 2014

    2014 arXiv

  5. [13]

    Deep residual learning for image recognition,

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

    2016

  6. [14]

    Densely connected convolutional networks,

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger, “Densely connected convolutional networks,” Proceedings of the IEEE conference on computer vision and pattern recognition , pp. 4700–4708, 2017

    2017

  7. [15]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” Proceedings of the IEEE/CVF International Conference on Computer Vision , pp. 10012–10022, 2021

    2021

  8. [16]

    Efficiently modeling long sequences with structured state spaces,

    Albert Gu, Karan Goel, and Christopher Re, “Efficiently modeling long sequences with structured state spaces,” International Conference on Learning Representations, 2021

    2021

  9. [17]

    Trans4mer: State-space models for vision,

    Thanh Nguyen and Xinyu Chen, “Trans4mer: State-space models for vision,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  10. [18]

    U-mamba: Enhancing long- range dependency for biomedical image segmentation,

    Jun Ma, Feifei Li, and Bo Wang, “U-mamba: Enhancing long- range dependency for biomedical image segmentation,” arXiv preprint arXiv:2401.04722, 2024

    2024 arXiv

  11. [19]

    Vision mamba: Efficient visual representation learning with bidirectional state space model,

    Lianghui Zhu, Bencheng Liao, Qian Zhang, Xinlong Wang, Wenyu Liu, and Xinggang Wang, “Vision mamba: Efficient visual representation learning with bidirectional state space model,” in Forty-first International Conference on Machine Learning

  12. [20]

    Vision mamba: A scalable state-space model for vision tasks,

    Jian Li, Ming Xu, and Fei Yang, “Vision mamba: A scalable state-space model for vision tasks,” arXiv preprint arXiv:2301.01234 , 2023

    2023 arXiv

  13. [21]

    Finding structure in time,

    Jeffrey L Elman, “Finding structure in time,” Cognitive Science, vol. 14, no. 2, pp. 179–211, 1990

    1990

  14. [22]

    Learning long-term dependencies with gradient descent is difficult,

    Yoshua Bengio, Patrice Simard, and Paolo Frasconi, “Learning long-term dependencies with gradient descent is difficult,” IEEE transactions on neural networks, vol. 5, no. 2, pp. 157–166, 1994

    1994

  15. [23]

    Long short-term memory,

    Sepp Hochreiter and J ¨urgen Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  16. [24]

    Efficient rnns with diagonal recurrence,

    Axel Orvieto, Krzysztof Choromanski, and Alexandros Papageor- giou, “Efficient rnns with diagonal recurrence,” arXiv preprint arXiv:2305.09029, 2023

    2023 arXiv

  17. [25]

    Complex diagonal recurrence and exponential gating for rnns,

    Laura Beck and Chris Thompson, “Complex diagonal recurrence and exponential gating for rnns,” Journal of Machine Learning Research , 2024

    2024

  18. [26]

    Minrnns: Simplified recurrent models for efficient sequence modeling,

    Liwei Feng, Wei Zhang, and Jisoo Kim, “Minrnns: Simplified recurrent models for efficient sequence modeling,” Neural Information Processing Systems, 2024

    2024

  19. [27]

    Imagenet: A large-scale hierarchical image database,

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei, “Imagenet: A large-scale hierarchical image database,” IEEE conference on computer vision and pattern recognition , pp. 248–255, 2009

    2009

  20. [28]

    Training data-efficient image transformers & distillation through attention,

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv ´e J ´egou, “Training data-efficient image transformers & distillation through attention,” arXiv preprint arXiv:2012.12877, 2021

    2012 arXiv

  21. [29]

    Decoupled weight decay regulariza- tion,

    Ilya Loshchilov and Frank Hutter, “Decoupled weight decay regulariza- tion,” arXiv preprint arXiv:1711.05101 , 2017

    2017 arXiv

  22. [30]

    Acceleration of stochastic approximation by averaging,

    Boris T Polyak and Anatoli B Juditsky, “Acceleration of stochastic approximation by averaging,” SIAM journal on control and optimization , vol. 30, no. 4, pp. 838–855, 1992

    1992

  23. [31]

    Aggregated residual transformations for deep neural networks,

    Saining Xie, Ross Girshick, Piotr Doll ´ar, Zhuowen Tu, and Kaiming He, “Aggregated residual transformations for deep neural networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1492–1500

    2017

  24. [32]

    Designing network design spaces,

    Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Doll ´ar, “Designing network design spaces,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 10428–10436, 2020

    2020

  25. [33]

    Upernet: Unified perceptual parsing for scene understanding,

    Tete Xia, Jianping Wang, Yilun Chen, Zhi Zhang, Dahua Lin, and Gang Yu, “Upernet: Unified perceptual parsing for scene understanding,” European Conference on Computer Vision , pp. 497–513, 2020

    2020

  26. [34]

    Semantics-aware adaptive knowledge distillation for sensor-to-vision action recognition,

    Yang Liu, Keze Wang, Guanbin Li, and Liang Lin, “Semantics-aware adaptive knowledge distillation for sensor-to-vision action recognition,” IEEE Transactions on Image Processing , vol. 30, pp. 5573–5588, 2021

    2021

  27. [35]

    Tcgl: Temporal contrastive graph for self-supervised video representation learning,

    Yang Liu, Keze Wang, Lingbo Liu, Haoyuan Lan, and Liang Lin, “Tcgl: Temporal contrastive graph for self-supervised video representation learning,” IEEE Transactions on Image Processing , vol. 31, pp. 1978– 1993, 2022

    1978

  28. [36]

    Enhanced soft label for semi-supervised semantic segmentation,

    Jie Ma, Chuan Wang, Yang Liu, Liang Lin, and Guanbin Li, “Enhanced soft label for semi-supervised semantic segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 1185–1195

    2023

  29. [37]

    Causal reasoning meets visual representation learning: A prospective study,

    Yang Liu, Yu-Shen Wei, Hong Yan, Guan-Bin Li, and Liang Lin, “Causal reasoning meets visual representation learning: A prospective study,” Machine Intelligence Research, vol. 19, no. 6, pp. 485–511, 2022

    2022

  30. [38]

    Skeletonmae: graph-based masked autoencoder for skeleton sequence pre-training,

    Hong Yan, Yang Liu, Yushen Wei, Zhen Li, Guanbin Li, and Liang Lin, “Skeletonmae: graph-based masked autoencoder for skeleton sequence pre-training,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2023, pp. 5606–5618

    2023

  31. [39]

    Hybrid-order representation learning for electricity theft detection,

    Yuying Zhu, Yang Zhang, Lingbo Liu, Yang Liu, Guanbin Li, Mingzhi Mao, and Liang Lin, “Hybrid-order representation learning for electricity theft detection,” IEEE Transactions on Industrial Informatics , vol. 19, no. 2, pp. 1248–1259, 2022

    2022

  32. [40]

    Cross-modal causal intervention for medical report generation,

    Weixing Chen, Yang Liu, Ce Wang, Jiarui Zhu, Shen Zhao, Guanbin Li, Cheng-Lin Liu, and Liang Lin, “Cross-modal causal intervention for medical report generation,” arXiv preprint arXiv:2303.09117 , 2023

    2023 arXiv

  33. [41]

    Aligning cyber space with physical world: A comprehensive survey on embodied ai,

    Yang Liu, Weixing Chen, Yongjie Bai, Xiaodan Liang, Guanbin Li, Wen Gao, and Liang Lin, “Aligning cyber space with physical world: A comprehensive survey on embodied ai,” arXiv preprint arXiv:2407.06886, 2024

    2024 arXiv

  34. [42]

    Towards long-horizon vision-language navigation: Platform, benchmark and method,

    Xinshuai Song, Weixing Chen, Yang Liu, Weikai Chen, Guanbin Li, and Liang Lin, “Towards long-horizon vision-language navigation: Platform, benchmark and method,” arXiv preprint arXiv:2412.09082 , 2024

    2024 arXiv

  35. [43]

    Fine-grained spatial-temporal mlp architecture for metro origin- destination prediction,

    Yang Liu, Binglin Chen, Yongsen Zheng, Guanbin Li, and Liang Lin, “Fine-grained spatial-temporal mlp architecture for metro origin- destination prediction,” arXiv preprint arXiv:2404.15734 , 2024

    2024 arXiv

  36. [44]

    Diversity matters: User-centric multi-interest learning for conversational movie recommendation,

    Yongsen Zheng, Guohua Wang, Yang Liu, and Liang Lin, “Diversity matters: User-centric multi-interest learning for conversational movie recommendation,” in Proceedings of the 32nd ACM International Conference on Multimedia , 2024, pp. 9515–9524

    2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.