Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Vision Mamba Distillation for Low-resolution Fine-grained Image Classification

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Mamba-based distillation sets state-of-the-art accuracy on seven low-resolution fine-grained benchmarks with a 7M-parameter student network.

desk verdict A useful but somewhat overclaimed empirical paper: the Mamba-student + SRGAN + multi-level distillation combo works, but test-set-selected hyperparameters and a confounded SOTA comparison mean the headline numbers should be read with care. read the letter →

arxiv 2411.17980 v1 pith:TIWIAGHU submitted 2024-11-27 cs.CV

Yao Chen , Jiabao Wang , Peichao Wang , Rui Zhang , Yang Li This is my paper · ORCID
classification cs.CV
keywords fine-grainedimageclassificationlow-resolutionknowledgedistillationMambastate-spacemodelsuper-resolutionefficiency
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

Low-resolution images lose the fine details needed to tell similar species, car models, or dog breeds apart, and existing fixes—super-resolution or knowledge distillation—roughly double the model size and compute. This paper proposes ViMD, a distillation scheme in which a tiny Mamba state-space student network (6.99M parameters, 0.50G FLOPs) is trained on low-resolution images under supervision from a high-resolution teacher of the same architecture. The teacher's knowledge is transferred through both a softened logit-matching loss and a layer-wise hidden-state matching loss. On seven public fine-grained benchmarks, the resulting student claims the best Top-1 accuracy of all compared methods—for instance 80.19% on CUB and 94.03% on Flowers—while keeping inference cost far below prior SR- and KD-based approaches. The paper's underlying message is that a carefully distilled state-space model can make low-resolution fine-grained classification both more accurate and more practical for embedded devices.

What carries the argument

The load-bearing component is the multi-level Mamba knowledge distillation loss $L_{MKD} = L_{LD} + \beta L_{HSD}$, where $L_{LD}$ is the KL divergence between the student's and teacher's softened logits and $L_{HSD} = \sum_{i=1}^{N} \|H^t_i - H^s_i\|_2^2$ aligns the hidden states of all $N$ Vision Mamba encoder layers. The student SRVM-Net first upscales a $56\times56$ low-resolution image with a frozen pre-trained SRGAN generator, then classifies the reconstructed image with Vim-Tiny; the teacher HRVM-Net is the same Vim-Tiny trained directly on $224\times224$ high-resolution images. Matching hidden states at every layer is what transfers the teacher's fine-grained feature geometry to the student, while the logit term keeps the student's decision boundary aligned with the teacher's.

What would settle it

Train and test the same ViMD pipeline on low-resolution images degraded with realistic blur, noise, and JPEG compression (or on a real low-resolution capture dataset) and compare Top-1 accuracy with the bicubic-downsampled results; if the accuracy drops substantially, the headline numbers are specific to the synthetic degradation model rather than to low-resolution classification in general.

Watch

Extended reading notes

Core claim

The central claim is that a lightweight student network built on the Vision Mamba (Vim-Tiny) backbone, fed super-resolved images from a frozen SRGAN generator, can close most of the accuracy gap against a high-resolution teacher when trained with a multi-level Mamba knowledge distillation loss. The loss combines the KL divergence of the teacher's and student's logits at temperature $\Delta = 4$ with the sum over layers of the L2 distances between their hidden states, added to the usual cross-entropy term. With this setup, the paper reports Top-1 accuracies of 80.19%, 88.93%, 84.18%, 92.56%, 94.03%, 78.43%, and 83.66% on CUB, Stanford Cars, Stanford Dogs, Oxford-IIIT Pet, Oxford-102 Flower, MIT67, and Stanford 40 Actions, respectively, the best among all compared methods, including SRKD and JSC with two different SR backbones. The same student has only 6.99M parameters and 0.50G FLOPs, roughly 60% of ResNet18's parameters and 27% of its FLOPs, and at test time only the student runs, so inference cost stays low.

Load-bearing premise

The reported state-of-the-art accuracy depends on the assumption that low-resolution images produced by bicubic downsampling of high-resolution images faithfully represent real low-resolution captures, so the gains may not transfer to images with genuine noise, blur, or compression artifacts.

Editorial extensions

If this is right

  • If the reported numbers hold, low-resolution fine-grained classification no longer requires the heavy CNN or Transformer backbones used by earlier SR and KD methods; a 7M-parameter Mamba student suffices to beat them.
  • The teacher is removed at test time, so the inference pipeline is just the frozen SRGAN generator plus Vim-Tiny, making the method suitable for embedded and mobile deployment.
  • The layer-wise hidden-state alignment transfers more than final predictions: it also transfers internal feature representations, which may improve generalization beyond what logit-only distillation would achieve.
  • The recipe generalizes beyond this particular backbone: the authors note Vim-Small and VMamba are drop-in alternatives, so the distillation loss is not tied to Vim-Tiny.

Reading between the lines

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

  • If the method's success carries over to real low-resolution captures, a natural next test is to evaluate on images with natural blur, noise, and compression instead of bicubic downsampling; the reported margins may shrink or survive depending on how well the SRGAN generator handles those degradations.
  • The hidden-state matching at every layer could be viewed as a strong feature-space regularizer; a perturbation study that corrupts SR outputs after training would show whether the student has actually learned robust fine-grained features or has memorized the specific bicubic degradation.
  • Because the best balance $\beta=20$ was chosen from test-set accuracy across four datasets, a retuning on a held-out validation split would provide a stricter, out-of-sample estimate of the method's true improvement.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript proposes ViMD, a knowledge-distillation method for low-resolution fine-grained image classification. The student network (SRVM-Net) consists of a frozen SRGAN generator followed by a Vim-Tiny classification sub-network; the teacher (HRVM-Net) is a Vim-Tiny network trained on high-resolution images. A multi-level distillation loss combines KL-divergence logit distillation with L2 hidden-state distillation. Experiments on seven fine-grained datasets report higher top-1 accuracy than four prior methods, together with lower parameter and FLOP counts for the deployed student. Ablations attribute the gains to the Vim-Tiny backbone and to the two distillation losses.

Significance. The proposed system is practically appealing: it is lightweight, the architecture is simple, the code is publicly available, and the combination of frozen SRGAN with a Mamba classifier and logit-plus-hidden-state distillation is a plausible design. If the reported numbers survive a properly out-of-sample evaluation, the paper would be a useful empirical contribution to low-resolution fine-grained classification. The strength of the claim, however, rests entirely on the seven top-1 accuracy numbers in Table I and on the ablation deltas in Table II, and those numbers are compromised by test-set-based hyperparameter selection and by the absence of repeated-run uncertainty estimates.

major comments (3)
  1. [Section III-C2, Figure 3, Table I] The central claim that ViMD achieves the best accuracy on all seven datasets is not strictly out-of-sample because the distillation balance beta was selected using test-set accuracy. In Section III-C2 the authors sweep beta in {1, 10, 20, 30}, report accuracies on the test sets of CUB, CAR, Action, and Flower, and recommend beta=20 because it achieves the best accuracies on CAR and Action; the same beta is then used to produce all Table I and Table II results. Since the test set has already been used for model selection, the quoted margins are inflated. This concern is consequential: the margin over JSC(SwinIR) on CAR is only 0.69 percentage points (88.93 vs. 88.24). The authors should select beta on a held-out validation split, report all four beta values on all seven datasets, and restate the headline results after doing so.
  2. [Tables I and II] All reported accuracies appear to be single runs with no multiple seeds and no error bars. Several important differences are small enough that they could reverse with different random seeds: for example, the PET ablation gain from adding L_HSD is 0.27 percentage points in Table II, and the CAR advantage over JSC(SwinIR) is 0.69 percentage points in Table I. The authors should provide mean and standard deviation over at least three independent runs, or otherwise justify that single-run numbers are stable for the conclusions drawn.
  3. [Section III-A1] The evaluation is conducted exclusively on bicubic-downsampled high-resolution images, as the authors correctly acknowledge. This limits the claim that the method solves real-world low-resolution fine-grained classification, where blur, noise, compression, and other degradations are common. I recommend that the manuscript either add a small robustness experiment with additional degradations, or explicitly narrow the title and conclusion claims to the synthetic bicubic setting.
minor comments (5)
  1. [Section III-B] The text reports the DOG accuracy as 84.14%, while Table I and the abstract report 84.18%; these numbers must be aligned.
  2. [Section II-A2, Eqs. (4) and (9)] The indexing in the ViM encoder definition is inconsistent: the text says the output H_{i+1} is obtained from H_i, while Eq. (4) writes H_i = E_i(H_{i-1}), and Eq. (9) uses terms with subscript i-1 to compute the output of layer i. Please rewrite the index convention carefully so the layer recursion is unambiguous.
  3. [Table I] The JSC(SwinIR) row lacks entries in the Params and FLOPs columns even though the efficiency comparison is an important part of the contribution; either provide the values or state explicitly that they are the same as JSC(SRGAN).
  4. [Table I] The phrasing 'all the best results' is slightly stronger than what the comparison supports: DRE-Net and DME-Net are evaluated only on CUB and CAR, and SRKD only on CUB. Only JSC is compared on all seven datasets. The table caption or text should make the dataset coverage explicit.
  5. [Full text] The rendered text contains apparent encoding artifacts, such as long '/uni000000...' strings in the author affiliation line and in the Figure 3 caption. These should be cleaned before publication.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline SOTA accuracies are partly constructed by test-set hyper-parameter selection: β=20 is chosen because it gives the best test accuracies on CAR and Action, and the same test accuracies are then reported as the method's best results.

  1. fitted input called prediction [Section III-C2 (Analysis of Hyper-parameters), Fig. 3; applied to Section III-B / Table I]
    "We recommend β as 20, because it achieves the best accuracies on CAR and Action. Although it does not achieve the best accuracies on CUB and Flower, the differences between it and the best accuracies are only 0.09% and 0.01%, and it still outperforms other SOTA method on all four datasets by 2.26%, 0.55%, 7.02% and 10.56%."

    The distillation-balance hyper-parameter β is selected using the test-set accuracies of CUB, CAR, Action, and Flower (Fig. 3), and β=20 is chosen because it achieves the best accuracies on CAR and Action. The same β=20 is then used to produce the headline results in Table I, where the paper claims 'all the best results' (e.g., 88.93% on CAR and 83.66% on Action). For those datasets, the reported 'prediction' is the very test accuracy that served as the selection objective: saying β=20 'achieves the best accuracies on CAR and Action' is just restating the argmax criterion used to pick β. The reported SOTA numbers are therefore fitted selection values rather than independent out-of-sample predictions.

full rationale

The paper's core method is an empirical combination of an externally pretrained SRGAN generator, Vim-Tiny, and standard distillation losses (KL logit distillation plus L2 hidden-state regression). These components are justified by external references (SRGAN, Vim, Hinton KD, TinyBERT) and are not derived from the paper's own claims, so there is no self-definitional or uniqueness-imported circularity in the architecture itself. The only significant circularity concern is the selection of the distillation-balance hyper-parameter β: Section III-C2 chooses β=20 by inspecting test-set accuracies on four datasets (Fig. 3), and Section III-B then reports those same test accuracies as the 'all best results' of ViMD in Table I. For CAR and Action, the statement that β=20 'achieves the best accuracies' is exactly the argmax criterion used to pick β, so the headline numbers on those datasets are fitted inputs rather than out-of-sample predictions. The ablation gains in Table II are also computed with this test-selected β. The authors' prior work JSC [13] appears as a baseline but is not load-bearing, and the remaining derivation chain has independent content. Overall this is partial circularity through test-set peeking rather than a fully self-consistent derivation, so the score is 6.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical objects, forces, or entities. Its central claim rests on standard model architectures and losses, plus one domain assumption about synthetic LR data and one methodological assumption about single-run stability. The only dataset-tuned free parameter that the headline results depend on is beta, chosen on the test sets.

free parameters (3)
  • distillation balance beta = 20 (grid {1,10,20,30})
    Chosen in Section III-C2 by comparing test-set accuracy on CAR and Action; this hyperparameter is part of the final loss in Eq. (12) and affects the reported accuracy.
  • logit distillation weight alpha = 1
    Set by hand based on previous KD works [30,33,34], not tuned in this paper. Used in Eq. (15).
  • KD temperature Delta = 4
    Set by hand following prior distillation practice, used in Eq. (13).
assumptions (4)
  • domain assumption Bicubic downsampling of HR images produces a valid proxy for low-resolution fine-grained images
    Section III-A1 explains that LR images are generated by bicubic downsampling because no dedicated LR FGVC dataset exists. All seven benchmark evaluations depend on this proxy.
  • domain assumption Knowledge distillation via KL on logits and L2 on hidden states transfers HR knowledge to an LR student
    Adapted from Hinton [14] and TinyBERT [31]; the paper provides no theoretical guarantee, only empirical support from the ablation in Table II.
  • domain assumption Vim-Tiny is a suitable backbone for fine-grained classification
    The paper adopts Vim-Tiny [19] and cites related Mamba works [22-24] as evidence, but provides no direct controlled comparison to a CNN or Transformer of equal size under identical training.
  • ad hoc to paper Single-run accuracies with no seeds are treated as stable estimates
    All results in Tables I and II lack error bars; several distillation gains (for example 0.27% on PET in Table II) are small enough that run-to-run variation could change the conclusion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vision Mamba Distillation for Low-resolution Fine-grained Image Classification." pith.science (2026). https://pith.science/paper/TIWIAGHU

@misc{pith2026241117980,
  author       = {Pith},
  title        = {Pith review of: Vision Mamba Distillation for Low-resolution Fine-grained Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TIWIAGHU}},
  note         = {Machine review of arXiv:2411.17980}
}
read the original abstract

Low-resolution fine-grained image classification has recently made significant progress, largely thanks to the super-resolution techniques and knowledge distillation methods. However, these approaches lead to an exponential increase in the number of parameters and computational complexity of models. In order to solve this problem, in this letter, we propose a Vision Mamba Distillation (ViMD) approach to enhance the effectiveness and efficiency of low-resolution fine-grained image classification. Concretely, a lightweight super-resolution vision Mamba classification network (SRVM-Net) is proposed to improve its capability for extracting visual features by redesigning the classification sub-network with Mamba modeling. Moreover, we design a novel multi-level Mamba knowledge distillation loss boosting the performance, which can transfer prior knowledge obtained from a High-resolution Vision Mamba classification Network (HRVM-Net) as a teacher into the proposed SRVM-Net as a student. Extensive experiments on seven public fine-grained classification datasets related to benchmarks confirm our ViMD achieves a new state-of-the-art performance. While having higher accuracy, ViMD outperforms similar methods with fewer parameters and FLOPs, which is more suitable for embedded device applications. Code is available at https://github.com/boa2004plaust/ViMD.

Figures

Figures reproduced from arXiv: 2411.17980 by the authors.

Figure 1
Figure 1. Effectiveness and efficiency comparison between our ViMD and other [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An overview of ViMD, which is mainly composed of an SRVM-Net (student), an HRVM-Net (teacher), and a multi-level Mamba knowledge distillation [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Results of hyper-parameters analysis on the four datasets. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 29 canonical work pages

  1. [36]

    Low-resolution Visual Recognition via Deep Feature Distillation,

    M. Zhu, et al. ,“Low-resolution Visual Recognition via Deep Feature Distillation,” in Proc. IEEE Int. Conf. Acoust. Speech. Signal Process. , 2019, pp. 3762-3766

    work page 2019

  2. [37]

    Feature map distillation of thin nets for low-resolution object recognition,

    M. Zhu, et al. ,“Feature map distillation of thin nets for low-resolution object recognition,” in IEEE Trans. Image Process. , 2022, pp. 1364- 1379

    work page 2022

  3. [38]

    Pixel Distillation: Cost-flexible Distillation across Image Sizes and Heterogeneous Networks,

    M. Zhu, et al.,“Pixel Distillation: Cost-flexible Distillation across Image Sizes and Heterogeneous Networks, ” IEEE Trans. Pattern Anal. Mach. Intell., pp. 1-15, 1 July. doi: 2024,10.1109/TPAMI.2024.3421277, [On- line]

    arXiv 2024

  4. [1]

    Fine-grained image analysis with deep learning: A survey,

    X. Wei, et al. , “Fine-grained image analysis with deep learning: A survey, ” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 44, no. 12, pp. 8927-8948, 1 Dec. 2022

    work page 2022

  5. [2]

    The caltech-ucsd birds-200-2011 dataset,

    C. Wah, et al., “The caltech-ucsd birds-200-2011 dataset,” 2011

    work page 2011

  6. [3]

    3d object representations for fine-grained categoriza- tion,

    J. Krause, et al., “3d object representations for fine-grained categoriza- tion,” in Proc. IEEE Int. Conf. Comput. Vis. Workshops , 2013, pp. 554- 561

    work page 2013

  7. [4]

    Novel dataset for fine-grained image categorization,

    A. Khosla, et al., “Novel dataset for fine-grained image categorization,” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit. Workshop, 2011

    work page 2011

  8. [5]

    Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network,

    C. Ledig, et al. ,“Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network,” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit., 2017, pp. 105-114

    work page 2017

Show all 40 references
  1. [6]

    Complemental Attention Multi-Feature Fusion Network for Fine-Grained Classification,

    Z. Miao, X. Zhao, J. Wang, Y . Li and H. Li, “Complemental Attention Multi-Feature Fusion Network for Fine-Grained Classification,” in IEEE Signal Process Lett. , vol. 28, pp. 1983-1987, 2021

    1983

  2. [7]

    SwinIR: Image Restoration Using Swin Transformer,

    J. Liang, et al., “SwinIR: Image Restoration Using Swin Transformer, ” in Proc. IEEE/CVF Int. Conf. Comput. Vis. Workshops , 2021, pp. 1833- 1844

    2021

  3. [8]

    Convolutional low-resolution fine-grained classification,

    D. Cai, et al., “Convolutional low-resolution fine-grained classification, ” in Pattern Recognit. Lett., vol. 119, pp. 166-171, Mar. 2019

    2019

  4. [9]

    Discriminative Feature Mining and Enhancement Net- work for Low-Resolution Fine-Grained Image Recognition,

    T. Yan, et al. , “Discriminative Feature Mining and Enhancement Net- work for Low-Resolution Fine-Grained Image Recognition, ” in IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 8, pp. 5319-5330, Aug. 2022

    2022

  5. [10]

    Discriminative information restoration and extraction for weakly supervised low-resolution fine-grained image recognition,

    T. Yan, et al., “Discriminative information restoration and extraction for weakly supervised low-resolution fine-grained image recognition, ” in Pattern Recognit., vol. 127, pp. 108629, Jul. 2022

    2022

  6. [11]

    Super-resolution guided knowledge distillation for low- resolution image classification,

    H. Chen, et al., “Super-resolution guided knowledge distillation for low- resolution image classification, ” in Pattern Recogn. Lett. , vol 155, pp. 62-68, Mar. 2022

    2022

  7. [12]

    Teaching where to look: Attention similarity knowledge distillation for low resolution face recognition,

    S. Shin, et al., “Teaching where to look: Attention similarity knowledge distillation for low resolution face recognition, ”in Proc. 17th Eur. Conf. Comput. Vis., 2022, pp. 631-647

    2022

  8. [13]

    Low-Resolution Armored Vehicle Identification Ap- proach via Joint Super-resolution and Knowledge Distillation,

    J. Wang, et al. , “Low-Resolution Armored Vehicle Identification Ap- proach via Joint Super-resolution and Knowledge Distillation, ” in Journal. Army Engineering University. PLA , vol 3, pp. 39-47, May

  9. [14]

    Distilling the Knowledge in a Neural Network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the Knowledge in a Neural Network,” 2015, arXiv: 1503.02531

    2015 arXiv

  10. [15]

    Deep Residual Learning for Image Recognition,

    K. He, et al. , “Deep Residual Learning for Image Recognition, ” in Proc. IEEE/CVF Int. Conf. Comput. Vis. , 2016, pp. 770-778

    2016

  11. [16]

    Very deep convolutional networks for large- scale image recognition,

    K. Simonyan, et al. , “Very deep convolutional networks for large- scale image recognition, ” in Proc. Int. Conf. Learn. Represent. , 2015, [Oneline]. Available: http://arxiv.org/abs/1409.1556

    2015 arXiv

  12. [17]

    ShuffleNet V2: Practical Guidelines for Effcient CNN Architecture Design,

    N. Ma, et al. , “ShuffleNet V2: Practical Guidelines for Effcient CNN Architecture Design, ” in Proc. 15th Eur. Conf. Comput. Vis., 2018, pp. 122-138

    2018

  13. [18]

    Mamba: Linear-Time Sequence Modeling with Selec- tive State Spaces,

    G. Albert, et al., “Mamba: Linear-Time Sequence Modeling with Selec- tive State Spaces,” 2023, arXiv: 2312.00752

    2023 arXiv

  14. [19]

    Vision Mamba: Efficient Visual Representation Learn- ing with Bidirectional State Space Model,

    L. Zhang, et al., “Vision Mamba: Efficient Visual Representation Learn- ing with Bidirectional State Space Model,” 2024, arXiv: 2401.09417

    2024 arXiv

  15. [20]

    VMamba: Visual State Space Model,

    Y . Liu, et al. , “VMamba: Visual State Space Model,” 2024, arXiv: 2401.10166

    2024 arXiv

  16. [21]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,

    A. Dosovitskiy, et al. , “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,” in Proc. Int. Conf. Learn. Represent. , 2021, [Online]. Available: https://openreview.net/forum?id=YicbFdNTTy

    2021

  17. [22]

    Res-VMamba: Fine-Grained Food Category Visual Classification Using Selective State Space Models with Deep Residual Learning,

    C. Chen, et al. , “Res-VMamba: Fine-Grained Food Category Visual Classification Using Selective State Space Models with Deep Residual Learning,” 2024, arXiv: 2402.15761

    2024 arXiv

  18. [23]

    MiM-ISTD: Mamba-in-Mamba for Efficient Infrared Small Target Detection,

    T. Chen, et al. , “MiM-ISTD: Mamba-in-Mamba for Efficient Infrared Small Target Detection,” 2024, arXiv: 2403.02148

    2024 arXiv

  19. [24]

    U-Mamba: Enhancing Long-range Dependency for Biomedical Image Segmentation,

    J. Ma, et al. , “U-Mamba: Enhancing Long-range Dependency for Biomedical Image Segmentation,” 2024, arXiv: 2401.04722

    2024 arXiv

  20. [25]

    Cats and dogs,

    O. M. Parkhi, et al., “Cats and dogs, ” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit., 2012, pp. 3498-3505

    2012

  21. [26]

    Automated Flower Classification over a Large Number of Classes,

    M. E. Nilsback, et al. , “Automated Flower Classification over a Large Number of Classes, ” in Proc. IEEE 6th Indian Conf. Comput Vis., Graph. Image Process., 2008, pp. 722-729

    2008

  22. [27]

    Recognizing indoor scenes,

    A. Quattoni and A. Torralba, “Recognizing indoor scenes, ” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit. , 2009, pp. 413-420

    2009

  23. [28]

    Human action recognition by learning bases of action attributes and parts,

    B. Yao, et al. , “Human action recognition by learning bases of action attributes and parts, ” in Proc. IEEE Int. Conf. Comput. Vis. , 2011, pp. 1331-1338

    2011

  24. [29]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,

    J. Devlin, et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, ” 2018, arXiv:1810.04805

    2018 arXiv

  25. [30]

    Mope-clip: Structured pruning for efficient vision- language models with module-wise pruning error metric,

    H. Lin, et al. ,“Mope-clip: Structured pruning for efficient vision- language models with module-wise pruning error metric,” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit. , 2024, pp. 27370-27380

    2024

  26. [31]

    Tinybert: Distilling bert for natural language understand- ing,

    X. Jiao, et al., “Tinybert: Distilling bert for natural language understand- ing, ” 2019, arXiv:1909.10351

    2019 arXiv

  27. [32]

    DynaBERT: Dynamic BERT with adaptive width and depth,

    H. Lu, et al. ,“DynaBERT: Dynamic BERT with adaptive width and depth,” in Adv. Neural Inf. Proces. Syst. , 2020, pp. 9782-9793

    2020

  28. [33]

    CLIP-KD: An Empirical Study of CLIP Model Distil- lation,

    C. Yang, et al.,“CLIP-KD: An Empirical Study of CLIP Model Distil- lation,” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit. , 2024, pp. 15952-15962

    2024

  29. [34]

    Class Attention Transfer Based Knowledge Distillation,

    Z. Guo, et al.,“Class Attention Transfer Based Knowledge Distillation,” in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recognit. , 2023, pp. 11868-11877

    2023

  30. [35]

    Cubic convolution interpolation for digital image pro- cessing,

    R. Keys, et al. ,“Cubic convolution interpolation for digital image pro- cessing,” in IEEE Trans. Acoust. Speech. Signal Process. , 1981, pp. 1153-1160

    1981

  31. [39]

    Effects of Image Degradation and Degradation Removal to CNN-Based Image Classification,

    Y . Pei, et al., “Effects of Image Degradation and Degradation Removal to CNN-Based Image Classification, ” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 4, pp. 1239-1253, 1 April. 2021

    2021

  32. [2024]

    doi: 10.12018/j.issn.2097-0730.20230505002, [Online]

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.