Pith. sign in

REVIEW 4 major objections 5 minor 56 references

Spatial frequency information fusion network for few-shot learning

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Adding low-frequency DCT structure to images improves few-shot classification.

desk verdict Simple DCT low-pass plus fusion as a few-shot preprocessing trick shows modest same-backbone gains, but the paper defines the central mask two incompatible ways and never pins down the method. read the letter →

arxiv 2506.18364 v1 pith:Z56Z6KYV submitted 2025-06-23 cs.CV

Wenqing Zhao , Guojia Xie , Han Pan , Biao Yang , Weichuan Zhang This is my paper · ORCID
classification cs.CV
keywords Few-shotlearningFrequencydomaininformationSpatialDatapreprocessingaugmentationClassificationperformanceDiscreteCosineTransformFine-grainedimage
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

The paper proposes SFIFNet, a network that feeds a few-shot classifier a modified input formed by adding an image's low-frequency Discrete Cosine Transform (DCT) reconstruction back onto the original image. The central claim is that this frequency-domain fusion gives the backbone, here a ResNet12 plus Vision Transformer, richer and more stable feature representations than using spatial pixels alone, and that this improves classification when labeled examples are scarce. The claim is supported mainly by an ablation in which the same backbone with the preprocessing outperforms the unmodified version by about 1.1 to 3.3 percentage points across a fine-grained bird dataset, a dog-breed dataset, and a custom animal dataset, in both 1-shot and 5-shot settings. If true, the method matters because it is a cheap input-level augmentation that can be dropped into existing few-shot classifiers without changing their training objectives.

What carries the argument

The load-bearing mechanism is a three-step preprocessing pipeline. First, the image is transformed with the 2D Discrete Cosine Transform, converting pixel values into frequency coefficients. Second, a low-pass mask keeps only the top-left $\lfloor h\sqrt{r}\rfloor \times \lfloor w\sqrt{r}\rfloor$ coefficients with retention ratio $r=0.15$ and zeroes the rest. Third, the inverse DCT reconstructs a low-frequency image that is added directly to the original pixels, and the sum becomes the network input. The paper also describes and rejects a gradient-threshold low-pass filter as too slow. This machinery is what carries the causal claim: any accuracy gain is attributed to the frequency-domain fusion rather than to a new classifier.

What would settle it

Run the same ResNet12+ViT backbone and training schedule on the fine-grained bird dataset while sweeping the retention ratio $r$ from 0.05 to 0.5 and varying the fusion weight from 0.25 to 2.0. If no retention ratio or weight beats the unmodified input by a margin outside the reported confidence intervals, or only a single narrow parameter value does, then the claimed benefit is not robust evidence for frequency fusion.

Watch

Extended reading notes

Core claim

SFIFNet's central discovery is that the low-frequency component of an image, recovered by taking a two-dimensional DCT, zeroing all coefficients outside the top-left block defined by $\lfloor h\sqrt{r}\rfloor \times \lfloor w\sqrt{r}\rfloor$ with $r=0.15$, and applying the inverse DCT, can be added pixel-wise to the original image to produce an input that yields better few-shot classification than the original image alone. The paper interprets this as integrating frequency-domain and spatial-domain information: the low-frequency reconstruction highlights main structure and suppresses high-frequency detail and noise, while the preserved original image keeps the details, so the fused input carries both. Across the three datasets, the fused input outperforms the unmodified input on the same ResNet12+ViT backbone in all six reported 5-way 1-shot and 5-way 5-shot settings, and SFIFNet also reports higher accuracy than the FRN, DeepEMD, DeepBDC, and BSNet baselines.

Load-bearing premise

The argument depends on the assumption that the low-frequency block of the DCT, fixed at 15% retention and added back additively, captures class-discriminative structure without destroying the high-frequency details needed to tell similar classes apart, a premise the paper does not test by varying the retention ratio or the fusion weight.

Editorial extensions

If this is right

  • Fusing the low-frequency DCT reconstruction into input images improves average classification accuracy on all three evaluated datasets in both 5-way 1-shot and 5-way 5-shot settings, by about 1.1 to 3.3 percentage points over the same ResNet12+ViT backbone.
  • Because the fusion happens before the backbone, the preprocessing can be inserted into existing few-shot pipelines as a drop-in input transform, as long as the DCT, IDCT, and addition preserve the image size.
  • Retaining only low-frequency structure suppresses high-frequency noise and detail variation, which the paper argues reduces the model's tendency to over-fit sample-specific noise when training data are scarce.
  • The method's gains on fine-grained benchmarks suggest that, for few-shot fine-grained classification, low-frequency structural and contour cues are a reliable source of discriminative signal.

Reading between the lines

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

  • The paper never varies the retention ratio $r=0.15$ or the fusion weight, so a natural next experiment is a sweep over both; if the gain holds over a wide range, the method is robust, and if it peaks only at a single narrow setting, the method is a tuned preprocessing trick rather than a general principle.
  • The headline comparison against FRN, DeepEMD, DeepBDC, and BSNet uses different backbones and different software versions, so the Table 2 margins are not a clean test; the same-backbone ablation in Table 3 is the comparison that actually isolates the contribution of frequency fusion.
  • A testable extension is to apply the same preprocessing to medical or camera-trap images where lesions or rare species are distinguished by local texture; if high-frequency cues are diagnostic, the current low-pass fusion may need to be replaced by a band-pass fusion that keeps some mid-frequency detail.
  • The paper's rejection of gradient-based low-frequency selection on speed grounds leaves open whether a better low-frequency mask could yield larger gains; the claim is about the fusion concept, not about the optimality of the square-corner mask.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a data-preprocessing method for few-shot image classification: apply a 2D DCT to each input image, zero out all coefficients above a low-frequency cutoff, reconstruct via IDCT, and add the reconstruction to the original image. The authors call the resulting network SFIFNet, using ResNet12 and ViT as a joint backbone. They evaluate on CUB-200-2011, Stanford Dogs, and a custom 30-class animal dataset under 5-way 1-shot and 5-way 5-shot tasks, report improved accuracy over FRN, BSNet, DeepEMD, and DeepBDC (Table 2), and provide a same-backbone ablation (Table 3) showing gains of roughly 1.1 to 3.3 points.

Significance. If the same-backbone gains in Table 3 are robust, the method offers a simple, computationally cheap input-level augmentation that could be dropped into many existing few-shot classifiers, and the idea of fusing frequency-domain structure with spatial detail is plausible for fine-grained recognition. The paper does not ship code or data, and because of the unresolved definitional ambiguity in Eq. (3) and the missing statistical analysis, the claimed gains cannot currently be attributed to the described preprocessing. The contribution is therefore potentially useful but not yet verifiable.

major comments (4)
  1. [Mathematical components, Eq. (3)] Eq. (3) defines the low-frequency mask as L_{i,j} = D_{i,j} when i < floor(h*sqrt(r)) and j < floor(w*sqrt(r)), which retains a square block of side approximately sqrt(r) of the image dimension, i.e. area r = 15% of all coefficients. The sentence immediately below Eq. (3) says the retained block is the top-left floor(hr) by floor(wr) region, which with r = 0.15 keeps only r^2 approximately 2.25% of coefficients. These two masks produce very different images, and every result in Tables 2 and 3 depends on which one is used. The paper must unambiguously state the mask shape, provide a sensitivity analysis over r, and ideally release the preprocessing code.
  2. [Ablation Studies, Table 3] Table 3 reports a single set of runs and uses the standard deviation over 1000 randomly generated tasks, not the variability across training runs. The reported gains of 1.1 to 3.3 percentage points are not accompanied by any significance test, multiple training seeds, or paired comparison, so they may be within run-to-run variability. The central claim needs at least five independent training runs with mean plus/minus standard deviation and a paired test, for example across the same sampled tasks.
  3. [Performance comparison, Table 2] Table 2 compares SFIFNet (ResNet12 plus ViT) against FRN, DeepEMD, and DeepBDC with ResNet12 and against BSNet with CNN, while the frameworks also differ (PyTorch 1.7.1 for SFIFNet and 2.2.0 for the others). These backbone and framework differences confound the comparison, so the large accuracy gaps cannot be attributed to the proposed frequency preprocessing. The only controlled evidence is Table 3; Table 2 should be explicitly framed as an overall system comparison, or all methods should be evaluated with the same backbone and training pipeline.
  4. [Materials and Methods / Results] The manuscript states that fused images are fed into 'a backbone composed of ResNet12 and ViT' and Figure 1 shows the pipeline, but it does not describe how ResNet12 and ViT are combined, which layers are used, how the two networks interact, or whether the same preprocessing is applied to support and query images. The fusion operation is also described only as 'directly add' with no specification of the fusion weight. Without these details the method is not reproducible and the experiments cannot be independently verified.
minor comments (5)
  1. [Abstract and Introduction] The manuscript contains numerous grammatical errors, e.g. 'which would effect the classification performance' should be 'affect', and the opening sentence is a fragment. A thorough language edit is needed.
  2. [Eq. (1) and Eq. (4)] The text says 'N and N are the width and height of the images respectively'; this should be 'N and M are the width and height'. The same inconsistency appears in the surrounding DCT/IDCT discussion.
  3. [Table 1] Table 1 has column headers 'Ntrain', 'Nval', and 'Ntest' without definitions; these should be clarified, and the captions of Tables 1, 2 and 3 should be formatted consistently.
  4. [Data availability statement] The data availability statement says 'All data generated or analysed during this study are included in this published article', but the custom animal dataset and the preprocessing code are not actually available. This statement should be revised to describe what is and is not released.
  5. [Figure 1] Figure 1 is referenced but is not described in sufficient detail; please enlarge it and annotate each component so the pipeline is understandable without the text.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the preprocessing is a fixed input-level transformation tested against an unmodified same-backbone baseline.

full rationale

The paper's central claim is that adding IDCT-reconstructed low-frequency DCT content to the original image improves few-shot classification. The supporting evidence is the Table 3 ablation, which compares the unmodified ResNet12+ViT backbone with the same backbone preceded by the proposed preprocessing. No equation defines the preprocessing in terms of the accuracies it is meant to predict: Eqs. (1)-(5) are standard DCT/IDCT formulas with an explicit low-frequency mask, and the fusion is described as directly adding the reconstructed image to the original. The retention ratio r=0.15 is a fixed hyperparameter chosen by the authors, not fitted to the test results, so the reported gains are not forced by construction. The self-citations, including the claim that Weichuan Zhang and his team first analyzed quantization-noise bias in few-shot fine-grained classification (ref. 39), are motivational and do not carry the derivation: the method's validity rests on the controlled empirical comparison, not on that prior claim. The inconsistency between Eq. (3)'s sqrt(r) mask and the surrounding text's hr x wr mask is a reproducibility and correctness ambiguity, but it is not a circular reduction because neither version of the mask is defined in terms of the predicted accuracies. I find no step in which a fitted parameter is renamed as a prediction, a uniqueness theorem is imported from the authors' prior work, or an ansatz is smuggled in via citation. Accordingly, the derivation is self-contained with respect to circularity, and the appropriate score is 0.

Assumptions & free parameters 2 free parameters · 2 assumptions · 0 invented entities

The free-parameter ledger includes the retention ratio and the implicit fusion weight, both chosen ad hoc without sensitivity analysis. The axioms are standard DCT math plus domain assumptions about low-frequency information being useful and additive fusion being safe. No new physical or conceptual entities are introduced; SFIFNet is a model configuration, not a postulated entity.

free parameters (2)
  • low_frequency_retention_ratio r = 0.15
    Eq. (3) retains DCT coefficients with indices below floor(h*sqrt(r)) and floor(w*sqrt(r)), while the text says the retained block is floor(h*r) by floor(w*r). With r=0.15 the two readings retain 15% versus 2.25% of coefficients. The paper fixes r with no sensitivity analysis.
  • fusion weight for low-frequency image = 1.0 (implicit)
    The fused input is the original image plus the IDCT reconstruction. No weighted fusion or learned combination is considered, and the influence of this choice on performance is not tested.
assumptions (2)
  • domain assumption DCT coefficient magnitude and classification-relevant structure are concentrated in low frequencies; high-frequency components are mostly noise and detail variation that can be discarded.
    This motivates zeroing all coefficients outside a top-left block in Eq. (3). The paper offers no per-dataset analysis showing which frequencies are discriminative for fine-grained categories.
  • ad hoc to paper Adding the low-frequency reconstructed image to the original image preserves or improves discriminative information.
    The pixel-wise addition in the preprocessing pipeline is introduced without derivation, without varying the fusion weight, and without feature-space analysis; the claimed gain depends on this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spatial frequency information fusion network for few-shot learning." pith.science (2026). https://pith.science/paper/Z56Z6KYV

@misc{pith2026250618364,
  author       = {Pith},
  title        = {Pith review of: Spatial frequency information fusion network for few-shot learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z56Z6KYV}},
  note         = {Machine review of arXiv:2506.18364}
}
read the original abstract

The objective of Few-shot learning is to fully leverage the limited data resources for exploring the latent correlations within the data by applying algorithms and training a model with outstanding performance that can adequately meet the demands of practical applications. In practical applications, the number of images in each category is usually less than that in traditional deep learning, which can lead to over-fitting and poor generalization performance. Currently, many Few-shot classification models pay more attention to spatial domain information while neglecting frequency domain information, which contains more feature information. Ignoring frequency domain information will prevent the model from fully exploiting feature information, which would effect the classification performance. Based on conventional data augmentation, this paper proposes an SFIFNet with innovative data preprocessing. The key of this method is enhancing the accuracy of image feature representation by integrating frequency domain information with spatial domain information. The experimental results demonstrate the effectiveness of this method in enhancing classification performance.

Figures

Figures reproduced from arXiv: 2506.18364 by the authors.

Figure 1
Figure 1. Flow of extracting low-frequency information in the frequency domain and performing information fusion Performance comparison Datasets have become one of the most critical roles in the development of FSFGIC, not only as a means for evaluating the classification accuracy of different FSFGIC methods, but also for greatly promoting the development of the field of FSFGIC (e.g., solving more complex, practical, and chall… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 49 canonical work pages

  1. [1]

    2.LeCun, Y ., Bengio, Y

    Ren, J.et al.Adaptive feature selection-based feature reconstruction network for few-shot learning.Available at SSRN 4962361. 2.LeCun, Y ., Bengio, Y . & Hinton, G. Deep learning.nature521, 436–444 (2015). 7/10

    work page 2015

  2. [3]

    & Zhang, W

    Liao, Y ., Gao, Y . & Zhang, W. Dynamic accumulated attention map for interpreting evolution of decision-making in vision transformer.Pattern Recognit.165, 111607 (2025)

    work page 2025

  3. [4]

    Mou, C.et al.Monitoring endangered and rare wildlife in the field: A foundation deep learning model integrating human knowledge for incremental recognition with few data and low cost.Animals13(2023)

    work page 2023

  4. [5]

    & Yuan, Y

    Chen, Y ., Guo, X., Pan, Y ., Xia, Y . & Yuan, Y . Dynamic feature splicing for few-shot rare disease diagnosis.Med. Image Analysis90, 102959, DOI: https://doi.org/10.1016/j.media.2023.102959 (2023)

    arXiv 2023

  5. [6]

    & Sun, C

    Zhang, W., Liu, X., Xue, Z., Gao, Y . & Sun, C. NDPNet: A novel non-linear data projection network for few-shot fine-grained image classification.arXiv preprint arXiv:2106.06988(2021)

    arXiv 2021

  6. [7]

    Tang, X.et al.Cascading attention enhancement network for rgb-d indoor scene segmentation.Comput. Vis. Image Underst.104411 (2025)

    work page 2025

  7. [8]

    Du, X.et al.CCL-MPC: Semi-supervised medical image segmentation via collaborative intra-inter contrastive learning and multi-perspective consistency.Neurocomputing621, 129287 (2025)

    work page 2025

  8. [9]

    & Zhang, W.-C

    Shui, P.-L. & Zhang, W.-C. Corner detection and classification using anisotropic directional derivative representations. IEEE Transactions on Image Process.22, 3204–3218 (2013)

    work page 2013

Show all 56 references
  1. [10]

    Li, Y ., Zhang, W. & Ji, L. Automotive radar mutual interference mitigation based on power-weighted hough transform in the time-frequency domain.IEEE Transactions on Veh. Technol.(2024)

    2024

  2. [11]

    Plasma Med

    Lei, T.et al.Semi-supervised 3d medical image segmentation using multi-consistency learning with fuzzy perception- guided target selection.IEEE Transactions on Radiat. Plasma Med. Sci.(2024)

    2024

  3. [12]

    & Sun, C

    Lu, J., Peng, G., Zhang, W. & Sun, C. Track-before-detect algorithm based on cost-reference particle filter bank for weak target detection.IEEE Access11, 121688–121701 (2023)

    2023

  4. [13]

    & Sun, C

    Jing, J., Liu, S., Wang, G., Zhang, W. & Sun, C. Recent advances on image edge detection: A comprehensive review. Neurocomputing503, 259–271 (2022)

    2022

  5. [14]

    & Sun, C

    Zhang, W. & Sun, C. Corner detection using second-order generalized gaussian directional derivative representations. IEEE transactions on pattern analysis machine intelligence43, 1213–1224 (2019)

    2019

  6. [15]

    & Zhou, Z.-F

    Zhang, W.-C., Wang, F.-P., Zhu, L. & Zhou, Z.-F. Corner detection using gabor filters.IET Image Process.8, 639–646 (2014)

    2014

  7. [16]

    & Zhang, W.-C

    Shui, P.-L. & Zhang, W.-C. Noise-robust edge detector combining isotropic and anisotropic gaussian kernels.Pattern Recognit.45, 806–820 (2012)

    2012

  8. [17]

    & Alshammari, N

    Zhang, W., Sun, C., Breckon, T. & Alshammari, N. Discrete curvature representations for noise robust image corner detection.IEEE Transactions on Image Process.28, 4444–4459 (2019)

    2019

  9. [18]

    & Sun, C

    Zhang, W. & Sun, C. Corner detection using multi-directional structure tensor with multiple scales.Int. J. Comput. Vis. 128, 438–459 (2020)

    2020

  10. [19]

    & Sun, C

    Jing, J., Gao, T., Zhang, W., Gao, Y . & Sun, C. Image feature information extraction for interest point detection: A comprehensive review.IEEE Transactions on Pattern Analysis Mach. Intell.45, 4694–4712 (2022)

    2022

  11. [20]

    & Zhang, W

    Li, Y . & Zhang, W. Traffic flow digital twin generation for highway scenario based on radar-camera paired fusion.Sci. reports13, 642 (2023)

    2023

  12. [21]

    & Gao, Y

    Zhang, W., Sun, C. & Gao, Y . Image intensity variation information for interest point detection.IEEE Transactions on Pattern Analysis Mach. Intell.45, 9883–9894 (2023)

    2023

  13. [22]

    & Sowmya, A

    Wang, M., Zhang, W., Sun, C. & Sowmya, A. Corner detection based on shearlet transform and multi-directional structure tensor.Pattern Recognit.103, 107299 (2020)

    2020

  14. [23]

    & Sun, C

    Lu, J., Zhang, W., Zhao, Y . & Sun, C. Image local structure information learning for fine-grained visual classification.Sci. Reports12, 19205 (2022)

    2022

  15. [24]

    & Sun, C

    Jing, J., Liu, C., Zhang, W., Gao, Y . & Sun, C. Ecfrnet: Effective corner feature representations network for image corner detection.Expert. Syst. with Appl.211, 118673 (2023)

    2023

  16. [25]

    Wang, M.et al.A principal component analysis-based feature optimization network for few-shot fine-grained image classification.Mathematics13, 1098 (2025)

    2025

  17. [26]

    & Zemel, R

    Snell, J., Swersky, K. & Zemel, R. Prototypical networks for few-shot learning.Adv. neural information processing systems30(2017). 8/10

    2017

  18. [27]

    Liao, Y ., Zhang, W., Gao, Y ., Sun, C. & Yu, X. Asrsnet: Automatic salient region selection network for few-shot fine-grained image classification. InInternational Conference on Pattern Recognition and Artificial Intelligence, 627–638 (Springer, 2022)

    2022

  19. [28]

    & Suk, H.-I

    Shen, D., Wu, G. & Suk, H.-I. Deep learning in medical image analysis.Annu. review biomedical engineering19, 221–248 (2017)

    2017

  20. [29]

    & Colantonio, S

    Pachetti, E. & Colantonio, S. A systematic review of few-shot learning in medical imaging.Artif. Intell. Medicine156, 102949, DOI: https://doi.org/10.1016/j.artmed.2024.102949 (2024)

    2024

  21. [30]

    Guo, Z., Ao, S. & Ao, B. Few-shot learning based oral cancer diagnosis using a dual feature extractor prototypical network. J. Biomed. Informatics150, 104584, DOI: https://doi.org/10.1016/j.jbi.2024.104584 (2024). 31.Chen, H.et al.Applying few-shot learning for in-the-wild cam...

    2024

  22. [32]

    Tools Appl.1–19 (2024)

    Tang, X.et al.Cascading context enhancement network for rgb-d semantic segmentation.Multimed. Tools Appl.1–19 (2024)

    2024

  23. [33]

    & Sun, C

    Zhang, W., Zhao, Y ., Gao, Y . & Sun, C. Re-abstraction and perturbing support pair network for few-shot fine-grained image classification.Pattern Recognit.148, 110158, DOI: https://doi.org/10.1016/j.patcog.2023.110158 (2024)

    2024

  24. [34]

    & Shen, C

    Zhang, C., Cai, Y ., Lin, G. & Shen, C. Deepemd: Few-shot image classification with differentiable earth mover’s distance and structured classifiers. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)(2020)

    2020

  25. [35]

    Xie, J., Long, F., Lv, J., Wang, Q. & Li, P. Joint distribution matters: Deep brownian distance covariance for few- shot classification. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 7972–7981 (2022)

    2022

  26. [36]

    Li, X.et al.Bsnet: Bi-similarity network for few-shot fine-grained image classification.IEEE Transactions on Image Process.30, 1318–1331 (2020)

    2020

  27. [37]

    & Qiu, Y

    Feng, N., Wang, F. & Qiu, Y . Novel approach for promoting the generalization ability of neural networks.Int. J. Signal Process.2, 131–135 (2006). 38.Pachetti, E. & Colantonio, S. A systematic review of few-shot learning in medical imaging (2024). 2309.11433

    2006 arXiv

  28. [39]

    & Zhang, W

    Wang, J., Lu, J., Yang, J., Wang, M. & Zhang, W. An unbiased feature estimation network for few-shot fine-grained image classification.Sensors24(2024)

    2024

  29. [40]

    J., De Sa, C

    Ratner, A. J., De Sa, C. M., Wu, S., Selsam, D. & Ré, C. Data programming: Creating large training sets, quickly.Adv. neural information processing systems29(2016)

    2016

  30. [41]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 1740–1749 (2020)

    Xu, K.et al.Learning in the frequency domain. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 1740–1749 (2020)

    2020

  31. [42]

    Singh, G., Mittal, A.et al.Various image enhancement techniques-a critical review.Int. J. Innov. Sci. Res.10, 267–274 (2014). 43.Khayam, S. A. The discrete cosine transform (dct): theory and application.Mich. State Univ.114, 31 (2003). 44.Ahmed, N., Natarajan, T. & Rao, K. R. ...

    2014

  32. [45]

    & Jalloh, I

    Aggoun, A. & Jalloh, I. Two-dimensional dct/idct architecture.IEE Proceedings-Computers Digit. Tech.150, 2–10 (2003). 46.Yang, L. & Jin, R. Distance metric learning: A comprehensive survey.Mich. State Universiy2, 4 (2006)

    2003

  33. [47]

    & Belongie, S

    Wah, C., Branson, S., Welinder, P., Perona, P. & Belongie, S. The caltech-ucsd birds-200-2011 dataset.California Inst. Technol.(2011)

    2011

  34. [48]

    & Li, F.-F

    Khosla, A., Jayadevaprakash, N., Yao, B. & Li, F.-F. Novel dataset for fine-grained image categorization: Stanford dogs. InProc. CVPR workshop on fine-grained visual categorization (FGVC), vol. 2 (2011)

    2011

  35. [49]

    & Zhang, W

    Wang, J., Lu, J., Yang, J., Wang, M. & Zhang, W. An unbiased feature estimation network for few-shot fine-grained image classification.Sensors24, 7737 (2024)

    2024

  36. [50]

    neural information processing systems29(2016)

    Vinyals, O., Blundell, C., Lillicrap, T., Wierstra, D.et al.Matching networks for one shot learning.Adv. neural information processing systems29(2016)

    2016

  37. [51]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020)

    Dosovitskiy, A. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020). 9/10

    2020 arXiv

  38. [52]

    Large-scale machine learning with stochastic gradient descent

    Bottou, L. Large-scale machine learning with stochastic gradient descent. InProceedings of COMPSTAT’2010: 19th International Conference on Computational StatisticsParis France, August 22-27, 2010 Keynote, Invited and Contributed Papers, 177–186 (Springer, 2010)

    2010

  39. [53]

    & Hertz, J

    Krogh, A. & Hertz, J. A simple weight decay can improve generalization.Adv. neural information processing systems4 (1991)

    1991

  40. [54]

    & Cremers, D

    Kukaˇcka, J., Golkov, V . & Cremers, D. Regularization for deep learning: A taxonomy.arXiv preprint arXiv:1710.10686 (2017)

    2017 arXiv

  41. [55]

    Zhang, H., Xu, H., Tian, X., Jiang, J. & Ma, J. Image fusion meets deep learning: A survey and perspective.Inf. Fusion76, 323–336 (2021)

    2021

  42. [56]

    Zhang, W., Zhao, Y ., Breckon, T. P. & Chen, L. Noise robust image edge detection based upon the automatic anisotropic gaussian kernels.Pattern Recognit.63, 193–205, DOI: https://doi.org/10.1016/j.patcog.2016.10.008 (2017)

    2017 doi

  43. [57]

    Ter Braak, C. J. & Prentice, I. C. A theory of gradient analysis. InAdvances in ecological research, vol. 18, 271–317 (Elsevier, 1988)

    1988

  44. [58]

    & Sun, C

    Ren, J., Li, C., An, Y ., Zhang, W. & Sun, C. Few-shot fine-grained image classification: A comprehensive review.AI5, 405–425 (2024)

    2024

  45. [59]

    & Hariharan, B

    Wertheimer, D., Tang, L. & Hariharan, B. Few-shot classification with feature map reconstruction networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 8012–8021 (2021)

    2021

  46. [60]

    big Data8, 1–74 (2021)

    Alzubaidi, L.et al.Review of deep learning: concepts, cnn architectures, challenges, applications, future directions.J. big Data8, 1–74 (2021)

    2021

  47. [61]

    & Gao, Y

    Pan, Z., Zhang, W., Yu, X., Zhang, M. & Gao, Y . Pseudo-set frequency refinement architecture for fine-grained few-shot class-incremental learning.Pattern Recognit.155, 110686, DOI: https://doi.org/10.1016/j.patcog.2024.110686 (2024)

    2024

  48. [62]

    & Shui, P.-L

    Zhang, W.-C. & Shui, P.-L. Contour-based corner detection via angle difference of principal directions of anisotropic gaussian directional derivatives.Pattern Recognit.48, 2785–2797, DOI: https://doi.org/10.1016/j.patcog.2015.03.021 (2015). Acknowledgements (not compulsory) Th...

    2015 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.