Pith. sign in

REVIEW 4 major objections 6 minor 40 references

Empowering Vision Transformers with Multi-Scale Causal Intervention for Long-Tailed Image Classification

T0 review · 4 major / 6 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 The paper claims that TSCNet, a two-stage causal intervention framework, removes both semantic and distributional biases in long-tailed image classification and outperforms existing methods on ViT backbones.

desk verdict Real gains on long-tailed ViT, but the causal backdoor story does not hold up under inspection. read the letter →

arxiv 2505.08173 v1 pith:T4ILC32B submitted 2025-05-13 cs.CV

Xiaoshuo Yan , Zhaochuan Li , Lei Meng , Zhuang Qi , Wei Wu , Zixuan Li , Xiangxu Meng This is my paper · ORCID
classification cs.CV
keywords long-tailedimageclassificationvisiontransformercausalinferencebackdooradjustmentcounterfactualaugmentationlogitsbiascalibrationfine-grainedvisualrepresentationclassimbalance
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

Long-tailed image classifiers trained on imbalanced data tend to lock onto head-class cues and misclassify visually similar tail classes. The paper argues that existing causal long-tail methods, built for CNNs, fail to transfer to Vision Transformers because ViT's global feature representation hides the associations between fine-grained features and predictions. It proposes TSCNet, which first performs backdoor interventions at both patch and feature levels to stop the model leaning on class-agnostic background information, then constructs a counterfactual balanced distribution to calibrate the logits. If the claim holds, ViT-based classifiers can reduce the head-tail performance gap without sacrificing head-class accuracy, and the approach transfers across CNN and transformer backbones.

What carries the argument

The load-bearing object is a structural causal model over image $X$, semantic confounder $S$, data-distribution confounder $D$, and prediction $Y$, with two backdoor paths $X \leftarrow S \rightarrow Y$ and $X \leftarrow D \rightarrow Y$. TSCNet is the two-stage framework that intervenes on these paths: patch-level intervention stacks random confounder patch embeddings with the image tokens, feature-level intervention uses a k-means++ prototype dictionary of confounder features pooled through normalized weighted geometric mean, and counterfactual logits bias calibration generates tail samples by interpolating Fourier amplitude spectra under a per-class strength $L_c^e$ that rises or falls by 0.1 depending on whether the class accuracy clears a threshold. Together these components are what let the model remove semantic bias at the representation level and distribution bias at the logit level.

What would settle it

On CIFAR-100-LT, train TSCNet with the confounder dictionary replaced by random noise patches of the same shape while keeping all other settings fixed; if tail accuracy does not drop below the reported dictionary version, the dictionary's semantic content is not doing the causal work. A complementary check is to replace the single-sample approximation in Eq. (2) with the full dictionary average: if tail accuracy does not improve or match, the causal graph or the approximation is mis-specified.

Watch

Extended reading notes

Core claim

The central claim is that two distinct confounders drive long-tail errors—semantic factors such as background (S) and the imbalanced data distribution itself (D)—and that both can be neutralized with causal interventions suited to ViT. In the hierarchical causal representation learning stage, TSCNet masks out detected subjects, builds a confounder dictionary of class-agnostic patch images, and approximates the backdoor adjustment $P(Y|\mathrm{do}(X))$ by concatenating a randomly sampled confounder patch at the token level and by integrating over confounder prototypes at the feature level. In the counterfactual logits bias calibration stage, it perturbs the Fourier amplitude spectrum of tail-class images with an adaptively adjusted strength per class, creating a counterfactual balanced distribution that recalibrates the decision boundary. On CIFAR-100-LT with ratio 0.01, TSCNet with VPT reaches 0.887 overall accuracy and 0.830 tail accuracy, and on VireoFood-172 it reaches 0.875, both above the compared baselines, while head-class accuracy stays around 0.93.

Load-bearing premise

The method assumes that the background patches it masks out of images are exactly the class-irrelevant cues the model latches onto, and that looking at one randomly chosen background patch per image is enough to undo their influence; if the real spurious cues are not in those masks, the causal fix misses the true bias.

Editorial extensions

If this is right

  • Causal long-tail methods on ViT should intervene on representations, not only calibrate logits; the paper's comparisons show TDE and xERM gain little on ViT while TSCNet improves tail accuracy substantially.
  • Head-class performance need not be sacrificed to improve tail classes: TSCNet reports head accuracy around 0.93 while tail accuracy rises to 0.83 on CIFAR-100-LT ratio 0.01 and to 0.82 on VireoFood-172.
  • The approach is backbone-agnostic: the same two stages improve ResNet50, plain ViT, and VPT, suggesting the intervention mechanism transfers rather than depending on one architecture's training recipe.
  • Adaptive per-class counterfactual strength outperforms fixed perturbation strengths, so tuning augmentation intensity by observed class accuracy is part of the method's effect.

Reading between the lines

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

  • If the causal graph is right, the same two-stage intervention—background confounder dictionary plus counterfactual distribution—could be applied to other transformer-based vision tasks where background is a known confounder, such as scene classification or fine-grained recognition; the paper does not test this.
  • The one-sample approximation in Eq. (2) implies a testable scaling prediction: sampling several confounder patches per image or enlarging the dictionary should improve tail accuracy up to a saturation point; the paper reports only fixed dictionary settings.
  • Because the second stage perturbs only Fourier amplitude and keeps phase intact, the method implicitly claims phase carries the causally relevant class information; replacing amplitude interpolation with phase interpolation should degrade performance if that claim is true.
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 / 6 minor

Summary. The paper proposes TSCNet, a two-stage method for long-tailed image classification with Vision Transformer backbones. In the first stage (HCRL), the method extracts class-agnostic background patches from training images to form a confounder dictionary and applies what it calls backdoor interventions at the patch level and the global feature level to reduce semantic bias. In the second stage (CLBC), it performs Fourier-based counterfactual augmentation with an adaptively adjusted per-class perturbation strength to calibrate logit bias caused by the long-tailed data distribution. The paper reports results on CIFAR-100-LT at imbalance ratios 0.01, 0.02, and 0.1 and on VireoFood-172, claiming state-of-the-art accuracy, especially with the VPT backbone, and includes an ablation study plus attention and confusion visualizations.

Significance. If the causal interpretation were fully supported, the paper would offer a modular two-stage recipe that is orthogonal to many existing long-tailed methods and that appears to improve tail-class accuracy on ViT backbones. The strengths are the clear ablation design in Table 2, the evaluation on two long-tailed benchmarks, and the attempt to make the intervention mechanism applicable to Transformer-based features. The paper is also honest enough to report head-class and tail-class accuracies separately, so the reader can see that the gains are not merely a redistribution of overall error. However, the central causal claims are currently not established: the backdoor adjustment is replaced by a single-sample approximation without an inference-time expectation, and the confounder dictionary is constructed from the observed images themselves, which conflicts with the exogenous-confounder graph in Figure 3. These issues are load-bearing because the paper's contribution is specifically a causal debiasing method, not just a heuristic augmentation scheme.

major comments (4)
  1. [§4.2, Eq. (2)] The single-sample approximation in Eq. (2) is not equivalent to the backdoor adjustment in Eq. (1). The identity P(Y|do(X)) = sum_S P(Y|X,S)P(S) requires an expectation over S, but the implementation uses one randomly sampled confounder patch s_k per image, giving P(Y|f(E(X),E(s_k))). Training with a random s_k minimizes E_s log P(Y|X,s), whereas the backdoor-adjusted predictor is log E_s P(Y|X,s); these are different objectives. The paper does not state whether inference averages over multiple s_k, uses one sample, or omits the patch entirely. Unless the inference procedure is specified and shown to approximate the expectation, the improvements attributed to patch-level intervention in Table 2 are compatible with standard input augmentation and do not validate the causal reading.
  2. [§4.2, Confounder Dictionary and Figure 3] The confounder S is not exogenous as required by the graph S -> X, S -> Y. In §4.2, each dictionary entry s_i is obtained by detecting the main subject of x_i and applying a mask, so S is a deterministic function of the observed image X. With S = h(X), the back-door path X <- S -> Y is not a standard confounding structure, and the condition for backdoor adjustment using S is not met. The authors should either redefine S as a separate variable sampled from an image-independent distribution, or explicitly state that the construction is an approximation and justify why deterministic masks can stand in for the confounder distribution P(S).
  3. [§4.2, Eqs. (4)-(6)] The feature-level intervention is not derived. Eq. (4) asserts an NWGM approximation without specifying which network layer it applies to, and Eq. (5) replaces E_c[g(c)] by a weighted sum of prototype features, but g(c) is never defined and Eq. (6) uses an inconsistent indexing (1/N_i sum over N). It is therefore unclear how the learned weights mu_i correspond to P(S) or how the parameterized form in Eq. (5) represents the backdoor-adjusted conditional P(Y|X,S). Since the feature-level module contributes to the +F ablation gains in Table 2, this missing derivation is a load-bearing gap.
  4. [§5.1 and Table 1] The central empirical claim that TSCNet "outperforms existing methods" rests on single-run comparisons with no error bars, repeated seeds, or significance tests. The margins in the ablation study (e.g., 0.864 vs. 0.877 for +I+F+C vs. +I+F+C+R on CIFAR-100-ratio0.02) are of a size that can easily be run-to-run variation for ViT training. Please report mean and standard deviation over at least three runs, or otherwise provide a statistical justification that the reported gaps are stable.
minor comments (6)
  1. [§4.3, Eqs. (11)-(13)] The variable VLoL appears in Eq. (13) but is not defined, and the update rule text says the strength increases "from easy to difficult" while the equations increase L_e^c when accuracy is high and decrease it otherwise; please clarify the intended update and specify bounds on L_e^c.
  2. [§4.1] The text says "S andB acts as confounder," but B is never defined, and the notation in Figure 3 uses D for the distribution confounder; please make the notation consistent.
  3. [§4.2, Eq. (2)] The summation indices are inconsistent: the first expression uses 1/N sum over j, while the final expression uses a single sampled k; please rewrite the equation with consistent notation.
  4. [Table 1] The H2T row on VireoFood-172 reports Acc@h = 0.630 and Acc@t = 0.876, which is qualitatively different from all other rows and may be a typo; please verify the reported values.
  5. [§5.4 and §4.2] There are several typos and broken references: "Feature-level Invention model" in §5.4, "conterfactual" after Eq. (10), "Eq.6" in the sentence above Eq. (5), and the citation "[48]" is not expanded in the bibliography.
  6. [§1] The term "sparse mechanism shift (SMS)" is introduced in the Introduction but is never defined or used in the method sections; please either define it and connect it to the two-stage design or remove it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TSCNet's numbers come from external test-set comparisons; Eq. (2)'s single-sample approximation is a correctness concern, not a circular reduction.

full rationale

The paper's central claim is supported by accuracy on held-out CIFAR-100-LT and VireoFood-172 test sets, compared against unrelated baselines. No parameter is fitted to the reported test numbers and then renamed as a prediction. Eq. (1) is the standard backdoor adjustment formula, and Eq. (2) explicitly approximates the sum over confounders by a single random patch sample; whether this is an adequate causal estimator is a validity question, not a circular one. The confounder dictionary S is constructed from training images via detection/Grad-CAM masks and used as an additional model input, which may be inconsistent with the assumed causal graph S→X, but again this is a modeling/identification concern rather than a definitional reduction of the reported result to its inputs. The adaptive strength L_e^c is updated from training accuracy with a threshold, an optimization schedule rather than a fitted quantity reported as a prediction. The only same-group citations ([Li et al., 2024c] and [Wang et al., 2024c]) appear in evaluation-protocol and related-work context and are not load-bearing for the derivation. No uniqueness theorem, no ansatz smuggled in by self-citation, and no renaming of a known result presented as a derivation are exhibited. Therefore the derivation chain is not circular; the skeptical concerns about Eq. (2) belong to correctness risk, not circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The paper rests on a standard SCM with two confounders, plus heuristic approximations (random sampling, NWGM) and a per-class adaptive augmentation schedule controlled by a validation threshold. The main free parameters are not reported, and the invented SMS notion is undefined.

free parameters (5)
  • gamma
    Threshold in Eq. 12 that triggers increase or decrease of Le_c; not specified in the paper.
  • alpha_gf
    Weight of the feature consistency loss in Eq. 15; not specified.
  • confounder prototype dictionary size l
    Number of prototypes in Sp; not specified.
  • initial counterfactual strength Le_c
    Initial perturbation strength per class and update step 0.1 in Eq. 12-13; initial value not specified.
  • learning rates and epochs = 0.01 for CIFAR, 1e-4 to 5e-5 for VireoFood
    Training hyperparameters; standard but needed for reproduction.
assumptions (6)
  • standard math Backdoor adjustment formula P(Y|do(X)) = sum_S P(Y|X,S)P(S) is valid in the stated SCM.
    Used in Eq. 1 and 7; standard Pearl backdoor adjustment.
  • domain assumption S (semantic/background) and D (data distribution) are confounders that affect both image X and prediction Y.
    Section 4.1 defines the SCM; this is the paper's causal model, not proven.
  • domain assumption The confounder dictionary S built from detected subject masks and inverse transformations captures class-agnostic semantic confounders.
    Section 4.2, patch-level intervention; no evidence that background patches are the true confounders.
  • standard math NWGM approximation of expectation over confounders in deep networks holds.
    Eq. 4-5, borrowed from Xu 2015; standard in causal attention literature.
  • domain assumption Fourier amplitude interpolation preserves high-level semantics while perturbing low-level statistics.
    Section 4.3, Eq. 8-10; assumption from CIRL (Lv et al. 2022).
  • domain assumption Adaptive per-class adjustment of Le_c based on current accuracy improves calibration.
    Section 4.3, Eq. 11-13; heuristic without proof.
invented entities (1)
  • Sparse Mechanism Shift (SMS)
    purpose: Invoked to justify that the two stages can independently intervene on different biases.
    Mentioned in Sections 1 and 4.4 but never defined or referenced; appears to be an invented or undeveloped concept.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Empowering Vision Transformers with Multi-Scale Causal Intervention for Long-Tailed Image Classification." pith.science (2026). https://pith.science/paper/T4ILC32B

@misc{pith2026250508173,
  author       = {Pith},
  title        = {Pith review of: Empowering Vision Transformers with Multi-Scale Causal Intervention for Long-Tailed Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T4ILC32B}},
  note         = {Machine review of arXiv:2505.08173}
}
read the original abstract

Causal inference has emerged as a promising approach to mitigate long-tail classification by handling the biases introduced by class imbalance. However, along with the change of advanced backbone models from Convolutional Neural Networks (CNNs) to Visual Transformers (ViT), existing causal models may not achieve an expected performance gain. This paper investigates the influence of existing causal models on CNNs and ViT variants, highlighting that ViT's global feature representation makes it hard for causal methods to model associations between fine-grained features and predictions, which leads to difficulties in classifying tail classes with similar visual appearance. To address these issues, this paper proposes TSCNet, a two-stage causal modeling method to discover fine-grained causal associations through multi-scale causal interventions. Specifically, in the hierarchical causal representation learning stage (HCRL), it decouples the background and objects, applying backdoor interventions at both the patch and feature level to prevent model from using class-irrelevant areas to infer labels which enhances fine-grained causal representation. In the counterfactual logits bias calibration stage (CLBC), it refines the optimization of model's decision boundary by adaptive constructing counterfactual balanced data distribution to remove the spurious associations in the logits caused by data distribution. Extensive experiments conducted on various long-tail benchmarks demonstrate that the proposed TSCNet can eliminate multiple biases introduced by data imbalance, which outperforms existing methods.

Figures

Figures reproduced from arXiv: 2505.08173 by the authors.

Figure 1
Figure 1. The illustration of the proposed TSCNet. It removes se [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The example of Similar-FP and Non-similar-FP on (a), the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The Causal view of Long-tailed Image Classification. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Illustration of the proposed TSCNet. It contains two main stages: HCRL and CLBC. The former introduces class-independent [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Comparison of our adaptive adjustment of the enhance [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 8
Figure 8. Figure 8: The error confusion analysis of TSCNet, and the compari [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 34 canonical work pages

  1. [1]

    Cuda: Curriculum of data augmentation for long-tailed recognition

    [Ahn et al., 2022] Sumyeong Ahn, Jongwoo Ko, and Se- Young Yun. Cuda: Curriculum of data augmentation for long-tailed recognition. In The Eleventh International Conference on Learning Representations,

    work page 2022

  2. [8]

    Reslt: Residual learn- ing for long-tailed recognition

    [Cui et al., 2022] Jiequan Cui, Shu Liu, Zhuotao Tian, Zhisheng Zhong, and Jiaya Jia. Reslt: Residual learn- ing for long-tailed recognition. IEEE transactions on pat- tern analysis and machine intelligence , 45(3):3695–3706,

    work page 2022

  3. [9]

    Generalized para- metric contrastive learning

    [Cui et al., 2023] Jiequan Cui, Zhisheng Zhong, Zhuotao Tian, Shu Liu, Bei Yu, and Jiaya Jia. Generalized para- metric contrastive learning. IEEE Transactions on Pattern Analysis and Machine Intelligence,

    work page 2023

  4. [10]

    Lpt: Long-tailed prompt tuning for image classification

    [Dong et al., 2023] Bowen Dong, Pan Zhou, Shuicheng Yan, and Wangmeng Zuo. Lpt: Long-tailed prompt tuning for image classification. In ICLR,

    work page 2023

  5. [11]

    Global and local mix- ture consistency cumulative learning for long-tailed visual recognitions

    [Du et al., 2023] Fei Du, Peng Yang, Qi Jia, Fengtao Nan, Xiaoting Chen, and Yun Yang. Global and local mix- ture consistency cumulative learning for long-tailed visual recognitions. In CVPR, pages 15814–15823,

    work page 2023

  6. [12]

    Deep residual learning for image recog- nition

    [He et al., 2016] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recog- nition. In CVPR, pages 770–778,

    2016

  7. [13]

    Disentangling label distribution for long- tailed visual recognition

    [Hong et al., 2021] Youngkyu Hong, Seungju Han, Kwanghee Choi, Seokjun Seo, Beomsu Kim, and Buru Chang. Disentangling label distribution for long- tailed visual recognition. In CVPR, pages 6626–6636,

    work page 2021

  8. [15]

    Decoupling representation and classi- fier for long-tailed recognition

    [Kang et al., 2019] Bingyi Kang, Saining Xie, Marcus Rohrbach, Zhicheng Yan, Albert Gordo, Jiashi Feng, and Yannis Kalantidis. Decoupling representation and classi- fier for long-tailed recognition. In ICLR,

    work page 2019

Show all 40 references
  1. [16]

    Improving vi- sual prompt tuning by gaussian neighborhood minimiza- tion for long-tailed visual recognition

    [Li et al., 2024a] Mengke Li, Ye Liu, Yang Lu, Yiqun Zhang, Yiu-ming Cheung, and Hui Huang. Improving vi- sual prompt tuning by gaussian neighborhood minimiza- tion for long-tailed visual recognition. arXiv preprint arXiv:2410.21042,

    arXiv

  2. [18]

    Cross-modal causal relational reasoning for event-level vi- sual question answering

    [Liu et al., 2023] Yang Liu, Guanbin Li, and Liang Lin. Cross-modal causal relational reasoning for event-level vi- sual question answering. IEEE Transactions on Pattern Analysis and Machine Intelligence , 45(10):11624–11641,

    2023

  3. [19]

    Causality-inspired invariant representation learning for text-based person retrieval

    [Liu et al., 2024] Yu Liu, Guihe Qin, Haipeng Chen, Zhiy- ong Cheng, and Xun Yang. Causality-inspired invariant representation learning for text-based person retrieval. In AAAI, volume 38, pages 14052–14060,

    2024

  4. [20]

    Causality inspired representation learning for domain generalization

    [Lv et al., 2022] Fangrui Lv, Jian Liang, Shuang Li, Bin Zang, Chi Harold Liu, Ziteng Wang, and Di Liu. Causality inspired representation learning for domain generalization. In CVPR, pages 8046–8056,

    2022

  5. [21]

    Ex- ploring the limits of weakly supervised pretraining

    [Mahajan et al., 2018] Dhruv Mahajan, Ross Girshick, Vig- nesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, and Laurens Van Der Maaten. Ex- ploring the limits of weakly supervised pretraining. In ECCV, pages 181–196,

    2018

  6. [23]

    Cadet: a causal disentan- glement approach for robust trajectory prediction in au- tonomous driving

    [Pourkeshavarz et al., 2024] Mozhgan Pourkeshavarz, Jun- rui Zhang, and Amir Rasouli. Cadet: a causal disentan- glement approach for robust trajectory prediction in au- tonomous driving. In CVPR, pages 14874–14884,

    2024

  7. [24]

    Deit-lt: Distillation strikes back for vi- sion transformer training on long-tailed datasets

    [Rangwani et al., 2024] Harsh Rangwani, Pradipto Mon- dal, Mayank Mishra, Ashish Ramayee Asokan, and R Venkatesh Babu. Deit-lt: Distillation strikes back for vi- sion transformer training on long-tailed datasets. InCVPR, pages 23396–23406,

    2024

  8. [25]

    Balanced meta-softmax for long- tailed visual recognition

    [Ren et al., 2020] Jiawei Ren, Cunjun Yu, Xiao Ma, Haiyu Zhao, Shuai Yi, et al. Balanced meta-softmax for long- tailed visual recognition. Advances in neural information processing systems, 33:4175–4186,

    2020

  9. [26]

    Grad-cam: visual explanations from deep networks via gradient-based localization

    [Selvaraju et al., 2020] Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: visual explanations from deep networks via gradient-based localization. Inter- national journal of computer vision , 128:336–359,

    2020

  10. [27]

    Parameter- efficient long-tailed recognition

    [Shi et al., 2023] Jiang-Xin Shi, Tong Wei, Zhi Zhou, Xin-Yan Han, Jie-Jing Shao, and Yu-Feng Li. Parameter- efficient long-tailed recognition. arXiv preprint arXiv:2309.10019,

    2023 arXiv

  11. [28]

    Unbiased scene graph genera- tion via two-stage causal modeling

    [Sun et al., 2023] Shuzhou Sun, Shuaifeng Zhi, Qing Liao, Janne Heikkil¨a, and Li Liu. Unbiased scene graph genera- tion via two-stage causal modeling. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(10):12562– 12580,

    2023

  12. [29]

    Long-tailed classification by keeping the good and removing the bad momentum causal ef- fect

    [Tang et al., 2020] Kaihua Tang, Jianqiang Huang, and Han- wang Zhang. Long-tailed classification by keeping the good and removing the bad momentum causal ef- fect. Advances in Neural Information Processing Systems, 33:1513–1524,

    2020

  13. [30]

    Vl-ltr: Learning class- wise visual-linguistic representation for long-tailed visual recognition

    [Tian et al., 2022] Changyao Tian, Wenhai Wang, Xizhou Zhu, Jifeng Dai, and Yu Qiao. Vl-ltr: Learning class- wise visual-linguistic representation for long-tailed visual recognition. In ECCV, pages 73–91. Springer,

    2022

  14. [31]

    Learning imbalanced data with vision transformers

    [Xu et al., 2023] Zhengzhuo Xu, Ruikang Liu, Shuo Yang, Zenghao Chai, and Chun Yuan. Learning imbalanced data with vision transformers. In CVPR, pages 15793–15803,

    2023

  15. [32]

    Show, attend and tell: Neural image caption generation with visual attention

    [Xu, 2015] Kelvin Xu. Show, attend and tell: Neural image caption generation with visual attention. arXiv preprint arXiv:1502.03044,

    2015 arXiv

  16. [34]

    Context de- confounded emotion recognition

    [Yang et al., 2023] Dingkang Yang, Zhaoyu Chen, Yuzheng Wang, Shunli Wang, Mingcheng Li, Siao Liu, Xiao Zhao, Shuai Huang, Zhiyan Dong, Peng Zhai, et al. Context de- confounded emotion recognition. In CVPR, pages 19005– 19015,

    2023

  17. [35]

    Deep long-tailed learn- ing: A survey

    [Zhang et al., 2023] Yifan Zhang, Bingyi Kang, Bryan Hooi, Shuicheng Yan, and Jiashi Feng. Deep long-tailed learn- ing: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(9):10795–10816,

    2023

  18. [36]

    Bi-causal: Group activ- ity recognition via bidirectional causality

    [Zhang et al., 2024] Youliang Zhang, Wenxuan Liu, Danni Xu, Zhuo Zhou, and Zheng Wang. Bi-causal: Group activ- ity recognition via bidirectional causality. In CVPR, pages 1450–1459,

    2024

  19. [37]

    Bbn: Bilateral-branch network with cumulative learning for long-tailed visual recognition

    [Zhou et al., 2020] Boyan Zhou, Quan Cui, Xiu-Shen Wei, and Zhao-Min Chen. Bbn: Bilateral-branch network with cumulative learning for long-tailed visual recognition. In CVPR, pages 9719–9728,

    2020

  20. [38]

    Pareto deep long-tailed recognition: A conflict- averse solution

    [Zhou et al., 2024] Zhipeng Zhou, Liu Liu, Peilin Zhao, and Wei Gong. Pareto deep long-tailed recognition: A conflict- averse solution. In ICLR,

    2024

  21. [39]

    Cross-domain empirical risk min- imization for unbiased long-tailed classification

    [Zhu et al., 2022] Beier Zhu, Yulei Niu, Xian-Sheng Hua, and Hanwang Zhang. Cross-domain empirical risk min- imization for unbiased long-tailed classification. In AAAI, volume 36, pages 3589–3597,

    2022

  22. [40]

    Generalized logit adjustment: Calibrat- ing fine-tuned models by removing label bias in founda- tion models

    [Zhu et al., 2024] Beier Zhu, Kaihua Tang, Qianru Sun, and Hanwang Zhang. Generalized logit adjustment: Calibrat- ing fine-tuned models by removing label bias in founda- tion models. Advances in Neural Information Processing Systems, 36, 2024

    2024

  23. [2012]

    Learning imbalanced datasets with label-distribution-aware margin loss

    [Cao et al., 2019] Kaidi Cao, Colin Wei, Adrien Gaidon, Nikos Arechiga, and Tengyu Ma. Learning imbalanced datasets with label-distribution-aware margin loss. Ad- vances in neural information processing systems , 32,

    2019

  24. [2015]

    Causal attention for vision-language tasks

    [Yang et al., 2021] Xu Yang, Hanwang Zhang, Guojun Qi, and Jianfei Cai. Causal attention for vision-language tasks. In CVPR, pages 9847–9857,

    2021

  25. [2016]

    Transfer knowledge from head to tail: Uncertainty calibration under long-tailed distribution

    [Chen and Su, 2023] Jiahao Chen and Bing Su. Transfer knowledge from head to tail: Uncertainty calibration under long-tailed distribution. In CVPR, pages 19978–19987,

    2023

  26. [2018]

    Causal transportability for visual recognition

    [Mao et al., 2022] Chengzhi Mao, Kevin Xia, James Wang, Hao Wang, Junfeng Yang, Elias Bareinboim, and Carl V ondrick. Causal transportability for visual recognition. In CVPR, pages 7521–7531,

    2022

  27. [2019]

    Deep-based ingredient recognition for cooking recipe re- trieval

    [Chen and Ngo, 2016] Jingjing Chen and Chong-Wah Ngo. Deep-based ingredient recognition for cooking recipe re- trieval. In MM, pages 32–41,

    2016

  28. [2020]

    Scalable k-means++

    [Bahmani et al., 2012] Bahman Bahmani, Benjamin Mose- ley, Andrea Vattani, Ravi Kumar, and Sergei Vassilvitskii. Scalable k-means++. arXiv preprint arXiv:1203.6402 ,

    2012 arXiv

  29. [2021]

    Visual prompt tuning

    [Jia et al., 2022] Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning. In ECCV, pages 709–727. Springer,

    2022

  30. [2022]

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

    [Alexey, 2020] Dosovitskiy Alexey. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv: 2010.11929 ,

    2020 arXiv

  31. [2023]

    Class-balanced loss based on effective number of samples

    [Cui et al., 2019] Yin Cui, Menglin Jia, Tsung-Yi Lin, Yang Song, and Serge Belongie. Class-balanced loss based on effective number of samples. In CVPR, pages 9268–9277,

    2019

  32. [2024]

    Contextual debiasing for vi- sual recognition with causal mechanisms

    [Liu et al., 2022] Ruyang Liu, Hao Liu, Ge Li, Haodi Hou, TingHao Yu, and Tao Yang. Contextual debiasing for vi- sual recognition with causal mechanisms. In CVPR, pages 12755–12765,

    2022

Pith tools

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