Pith. sign in

REVIEW 6 major objections 6 minor 57 references

Did Models Learn Sufficiently? Attribution-Guided Training via Subset-Selected Counterfactual Augmentation

T0 review · 6 major / 6 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Training with counterfactual masks lifts ImageNet-100 to 91.1% and ImageNet-R to 62.6%

desk verdict A promising attribution-guided augmentation idea whose headline numbers don't match its own tables and whose label-preservation assumption is unvalidated; needs revision, not rejection. read the letter →

arxiv 2511.12100 v3 pith:Y4JGECS2 submitted 2025-11-15 cs.CV

Yannan Chen , Ruoyu Chen , Wei Wang , Bin Zeng , Jinke Li , Shiming Liu , Qunli Zhang , Yaowei Wang
show 1 more author
Xiaochun Cao
This is my paper
classification cs.CV
keywords counterfactualaugmentationattribution-guidedtrainingsubsetselectionshortcutlearningout-of-distributiongeneralizationmodelrobustnessimageclassification
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

Many image classifiers rely on a single decisive visual cue, making them fail when that cue is missing or the distribution shifts. This paper proposes SS-CA, an attribution-guided training strategy that feeds a model's own counterfactual explanations back into the training loop. Using a subset-selection-based counterfactual attribution, it finds the minimal set of regions whose removal flips the prediction to a competing class, replaces those regions with background, and trains the model jointly on original and augmented samples. The authors report consistent gains in in-distribution accuracy and out-of-distribution robustness across multiple ImageNet variants and backbone architectures, with ImageNet-100 accuracy rising from 89.50% to 91.14% and ImageNet-R from 60.94% to 62.59% on a contrastively pretrained visual encoder.

What carries the argument

The central object is a subset-selected counterfactual attribution computed by greedy submodular optimization. It seeks the minimal set of spatial regions whose removal pushes the model's confidence toward a counterfactual class while suppressing the ground-truth class, and whose insertion maintains fidelity to the original prediction. The augmentation then replaces these regions with background patches from a donor image, and a hard-mining filter retains only samples where the counterfactual flip confidence exceeds a threshold. Joint optimization on original and augmented batches with cross-entropy loss is what carries the training signal.

What would settle it

Measure human recognition accuracy on augmented samples (original image with the SS-CA mask replaced by background). If a sizable fraction are not recognized as the original class, the ground-truth labeling of these hard samples is unjustified, and the reported gains may partly reflect label noise rather than causal debiasing.

Watch

Extended reading notes

Core claim

The paper's central claim is that counterfactual attributions—explanations that identify which minimal region removals would change a model's prediction—can be used as training augmentations to correct incomplete causal learning. The proposed algorithm, SS-CA, extends a subset-selection attribution method into a counterfactual search that maximizes a utility function balancing the push toward a competing class and the fidelity to the ground-truth class. The resulting mask is applied to the image by replacing the selected regions with patches from a donor background image, and the augmented sample is given the original ground-truth label. Training on these hard samples together with the origi

Load-bearing premise

The method assumes that an image with its most decision-relevant region erased and replaced by arbitrary background is still recognizably the original class, and this label-preservation assumption is not human-validated.

Editorial extensions

If this is right

  • Counterfactual attribution can be more than a diagnostic tool; feeding it back into training actively steers the model toward more complete feature reliance.
  • Improved performance on renditions and sketches suggests reduced dependence on texture and style shortcuts.
  • Robustness to Gaussian noise and other corruptions improves without needing corruption-specific augmentation at test time.
  • The method requires no extra human annotations—only the model's own logits and background image patches.
  • The gains appear across convolutional, transformer, and contrastively pretrained encoders, indicating architecture generality.

Reading between the lines

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

  • The paper's labeling assumption—that an image with its most decision-relevant region replaced by background still belongs to the original class—could be tested directly by human recognition experiments; if a large fraction of augmented images are unrecognizable, the method may be inadvertently training on mislabeled examples, and the reported gains might partly reflect label noise rather than caus
  • Varying the source of donor background patches (e.g., same-class, other-class, or random natural images) could change the strength of the intervention and reveal with greater precision how the augmentation debiases the model.
  • The hard-mining threshold tau_aug controls a trade-off between the quantity and quality of augmented samples; a systematic sweep of this threshold could show whether the improvement is driven by the hardest samples or merely by the volume of extra data.
  • The counterfactual search cost scales with the number of region candidates per image; amortizing the search across epochs or using a faster approximation would make the method more practical for large-scale training.
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

6 major / 6 minor

Summary. The paper proposes SS-CA (Subset-Selected Counterfactual Augmentation), a training-time method that uses subset-selection-based counterfactual attribution (Counterfactual LIMA) to identify minimal image regions whose removal flips a model's prediction to a competing class. These regions are replaced with background patches from a donor image, and the resulting augmented sample is assigned the original ground-truth label and added to the training batch alongside the original samples. The authors claim that this closed-loop procedure reduces reliance on limited sufficient causes, improving in-distribution accuracy, out-of-distribution generalization, and robustness to corruptions across ImageNet-100, TinyImageNet-200, and ImageNet-1k, with several backbone architectures. The main results are reported in Tables 1--4, and ablations in Table 5 compare counterfactual LIMA with Grad-CAM and LIMA guidance.

Significance. If the claims are validated, the paper would make a useful contribution to explanation-guided training: it demonstrates a concrete mechanism for feeding faithful, subset-selected counterfactual attributions back into the training loop, and its ablation comparing factual versus counterfactual guidance (Table 5) is a sensible sanity check. The formulation in Eqs. (2)--(6) is explicit, and the use of background refilling rather than zero-masking is a reasonable design choice. However, the current evidence is not sufficient to support the central causal claim. The label-preservation assumption in Eq. (9) is load-bearing and unvalidated, key hyperparameters are omitted, and there are internal numerical inconsistencies, most notably between the abstract's claimed gains and the numbers in Table 3. The statistical reliability of the reported improvements is also unclear because no error bars or run counts are provided.

major comments (6)
  1. [Section 4.2, Eq. (9), Algorithm 1] The label-preservation assumption is the core inductive bias but is never validated. Hard mining retains only samples with c_max > τ_aug, i.e., images for which the model flipped away from y_gt after masking. Assigning these augmented images the original label y_j presupposes that the remaining visible content is semantically sufficient for y_gt. The paper asserts that humans can still recognize the object, but provides no human study, no oracle probe, and no label-validity control. This is especially problematic because the main results are interpreted as a causal debiasing effect rather than label-noise regularization or background-artifact learning. Please add a direct validation of the labeling assumption (e.g., human annotation of augmented samples, an oracle-based check, or an ablation with flipped/donor labels).
  2. [Abstract vs. Table 3] The abstract claims 'gains of 5.70%/18.04% on ImageNet-1k/ImageNet-R with CLIP ViT/32b', but Table 3 reports 72.07% vs. 71.44% ID (+0.63%) and 31.89% vs. 31.63% OOD (+0.26%) for the same setting. This is a large discrepancy that cannot be attributed to rounding. Please reconcile the abstract with the actual table, and if the 5.70%/18.04% figures come from a different configuration or dataset, state that clearly. As written, this inconsistency undermines confidence in the reported results.
  3. [Section 5.1, Tables 1--4] The text states that all scores are averaged over multiple runs 'to ensure statistical reliability', but no standard deviations, error bars, or number of runs are reported anywhere. Many gains are small (e.g., Table 3: +0.26% on ImageNet-R, +0.41% on ImageNet-S; Table 2 ViT: +0.52% ID). Without variance information it is impossible to determine whether these differences are significant. Please report means with standard deviations and, if feasible, significance tests or confidence intervals.
  4. [Section 4.1--4.2, Algorithm 1] The method depends on several hyperparameters that are never specified in the main text: λ1, λ2, the stopping threshold τ_cf, the hard-mining threshold τ_aug, the region budget k, the sub-region partition size m, and the donor-background patch selection procedure. The reader is referred to 'supplementary materials', but none are provided in the manuscript. These parameters directly control which samples are augmented and how aggressively, so the reported results cannot be reproduced or compared fairly without them. Please include the values and a sensitivity analysis.
  5. [Table 5 and Section 5.3] There are internal inconsistencies in the ablation results. The text says the full SS-CA framework 'achieves the peak performance of 91.18% (ID), 62.64% (ImageNet-R), and 59.11% (ImageNet-S)', but the 'w/ Additional Background' row in Table 5 reports 91.28, 62.68, 59.14, while the 'w/ Counterfactual LIMA' row reports 91.14, 62.59, 59.07. Section 5.2 also refers to 91.14/62.59/59.07 as the SS-CA result. Please correct the text or table so that the numbers are consistent.
  6. [Section 5.1, Tables 1--3] The experimental support for the central claim is thinner than the text suggests. On ImageNet-1k only one backbone (CLIP ViT-B/32, linear probing) is evaluated, while Tables 1--2 use three backbones. The comparison set is also narrow: only two recent debiasing baselines are included, and no standard augmentation methods (e.g., CutMix, RSC) or other explanation-guided augmentation baselines are compared, despite the related work surveying such methods. This limits the strength of the 'consistent improvement' claim and makes it hard to assess whether the gains are specific to counterfactual subset selection or shared by broader augmentation strategies. Please expand the baselines and, if feasible, add IN-1k results for at least one more backbone.
minor comments (6)
  1. [Global] The title and introductory sentence contain a grammatical error: 'Did Models Sufficient Learn?' should be 'Did Models Learn Sufficiently?'.
  2. [Figure 6] Figure 6 is referenced in Section 5.3, but the actual image content is not present in the manuscript; only the caption appears. Please include the figure or remove the reference.
  3. [Notation, Eq. (1)] The notation S_{·,j} for the cumulative subset is used without a formal definition. Please define it explicitly, e.g., S_{·,j} = {s_1, ..., s_j}.
  4. [References] Reference [37] for Tiny-ImageNet-200 uses an unusual URL format (cstr.cn). Please provide the standard citation or a stable DOI.
  5. [Section 1] The contributions section claims 'Extensive experiments on three ImageNet variants', while the abstract says 'five ImageNet variants'. The experimental section actually evaluates five datasets (IN-100, TinyImageNet-200, IN-1k, ImageNet-R, ImageNet-S). Please make this consistent.
  6. [Section 4.1, Eq. (6)] The four-term utility function is clear, but the role of the insertion score as a 'fidelity regularizer' could be made more precise. In particular, the insertion term rewards f_y_gt(I(S)) and penalizes f_y_counter(I(S)), but the paper does not discuss what happens when the selected regions are not actually the most discriminative for y_gt. A brief discussion or a synthetic example would help.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: SS-CA's reported gains are measured on held-out ID/OOD/corruption sets, and the self-cited LIMA component is ablated rather than assumed.

full rationale

The derivation chain runs from Eq. (2) (joint task-plus-augmentation loss), through Eq. (6) (counterfactual utility using model confidence), Algorithm 1 (greedy region search), Eq. (8) (background refilling), Eq. (9) (hard-batch labeling with original y_j), to Eq. (10) (joint optimization). Every headline result (Tables 1-4) is an external measurement on held-out ImageNet-100/TinyImageNet-200/ImageNet-1k ID sets and ImageNet-R/S OOD sets, so the gains are not forced by construction: the augmented-sample generator is not the evaluator. The method makes a load-bearing empirical assumption that the masked image I'_j still deserves label y_j (Sec. 4.2, Eq. 9); this is an unvalidated label-preservation assumption and a correctness risk, but it is not a circular derivation, because the reported ID/OOD accuracy is still an independent measurement of the trained model. The main self-referential element is reliance on the authors' LIMA [8] (and related work [6,7]) to justify 'faithful' subset attributions and to build Counterfactual LIMA. That reliance is not a uniqueness theorem and does not forbid alternatives; moreover, Table 5 ablates factual LIMA (90.41% ID) against Counterfactual LIMA (91.14% ID), so the central contribution is tested rather than reduced to the self-citation. No equation in the paper is equivalent by construction to its own input, and no fitted parameter is renamed as a prediction. Score 1 reflects only minor, non-load-bearing self-citation.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the label-preservation assumption for augmented samples, the asserted submodularity of the attribution objective, and several unreported hyperparameters. These are not derived or independently validated, and the missing hyperparameters and donor details make the method difficult to reproduce.

free parameters (6)
  • lambda_1, lambda_2 = not reported
    Weights in the counterfactual utility function (Eq. 4-6); no values or sensitivity analysis given in the main text.
  • tau_cf (stopping threshold) = not reported
    Stops greedy search when counterfactual confidence exceeds threshold (Algorithm 1, line 19).
  • tau_aug (hard mining threshold) = not reported
    Filters augmented samples whose max counterfactual confidence exceeds tau_aug (Eq. 9).
  • k (region budget) = not reported
    Maximum number of regions to remove in Algorithm 1; affects mask size.
  • sub-region partition (m) = not reported
    Image is partitioned into m disjoint regions (Section 3.1); m not specified.
  • background donor patch selection = n/a
    Donor images are 'curated, in-distribution collection of background patches derived from ImageNet and its variants' — construction details and matching criteria not specified.
assumptions (3)
  • domain assumption The utility function F(S) in Eq. 6 is submodular, so greedy selection is near-optimal.
    Section 4.1 calls it a submodular utility function but does not prove submodularity; neural network logits over subsets generally do not guarantee submodularity.
  • domain assumption An image with the counterfactual-masked region replaced by background remains an instance of the original class.
    The method assigns the original label to augmented samples (Eq. 9-10). This is the core intervention assumption; it is asserted ('humans can still easily recognize it') but not validated.
  • domain assumption The minimal decision-changing regions identified are the 'shortcuts' or spurious features; removing them is beneficial for robustness.
    The paper equates the model's most-decisive region with a limitation; this is the motivating claim but no causal experiment isolates that the removed regions are spurious.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Did Models Learn Sufficiently? Attribution-Guided Training via Subset-Selected Counterfactual Augmentation." pith.science (2026). https://pith.science/paper/Y4JGECS2

@misc{pith2026251112100,
  author       = {Pith},
  title        = {Pith review of: Did Models Learn Sufficiently? Attribution-Guided Training via Subset-Selected Counterfactual Augmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y4JGECS2}},
  note         = {Machine review of arXiv:2511.12100}
}
read the original abstract

Current visual models often make predictions based on a limited set of discriminative visual cues. As a result, they may become unreliable when the distribution shifts or when these cues are missing. Faithful attribution methods can reveal such problematic reliance through localized explanations, but they are typically used post hoc and are not fed back into the model. To address this limitation, we propose Subset-Selected Counterfactual Augmentation (SS-CA), a training strategy that masks decision-relevant regions to construct counterfactual samples and guide the model toward more robust decision boundaries. Specifically, we extend LIMA, a subset-selection-based faithful attribution method, to Counterfactual LIMA to identify regions whose removal shifts the model toward a competing class. SS-CA then selects near-boundary masks that reduce the logit gap while preserving the original semantics, and applies an adaptive counterfactual filling strategy to replace the masked regions without introducing external semantics. Feeding these counterfactual samples back into training encourages the model to exploit the remaining informative evidence and shifts the decision boundary toward a more robust one. Extensive experiments across five ImageNet variants show that SS-CA effectively improves ID accuracy, OOD generalization, and perturbation robustness, achieving gains of 5.70%/18.04% on ImageNet-1k/ImageNet-R with CLIP ViT/32b, 9.52%/11.33% on ImageNet-R/ImageNet-S on TinyImageNet-200 with ResNet-101, and about 4% under Gaussian Noise corruption. The code will be released soon.

Figures

Figures reproduced from arXiv: 2511.12100 by the authors.

Figure 1
Figure 1. Conceptual motivation for identifying and mitigating [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The overall framework of Subset-Selected Counterfactual Augmentation (SS-CA). It forms a closed training loop with three stages: (1) Conventional Training, where a factual image (e.g., “Robin”) is fed into the network to obtain an initial prediction; (2) Counterfactual Explanation, which identifies a minimal set of regions whose removal flips the prediction to a counterfactual class (e.g., “Heron”); and (3) Attribut… view at source ↗
Figure 3
Figure 3. Visualization of the SS-CA training loop on ImageNet [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Demonstrating SS-CA’s scalability during the training [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparative analysis of attribution guidance. Standard [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 1 linked inside Pith

  1. [1]

    Jamil Ahmad, Wail Gueaieb, Abdulmotaleb El Saddik, Giu- lia De Masi, and Fakhri Karray. Leveraging model explain- ability and fine-grained cutmix augmentation for robust de- tection of apricot diseases in uav images.Expert Systems with Applications, 296(1):128946:1–15, 2026. 3

    2026

  2. [2]

    Diffusion visual counterfactual explana- tions

    Maximilian Augustin, Valentyn Boreiko, Francesco Croce, and Matthias Hein. Diffusion visual counterfactual explana- tions. InAdvances in Neural Information Processing Sys- tems, pages 364–377, 2022. 2

    2022

  3. [3]

    Explaining image classifiers by counter- factual generation

    Chun-Hao Chang, Elliot Creager, Anna Goldenberg, and David Duvenaud. Explaining image classifiers by counter- factual generation. InInternational Conference on Learning Representations, pages 1–11, 2019. 2

    2019

  4. [4]

    Grad-cam++: General- ized gradient-based visual explanations for deep convolu- tional networks

    Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasubramanian. Grad-cam++: General- ized gradient-based visual explanations for deep convolu- tional networks. In2018 IEEE winter conference on appli- cations of computer vision (WACV), pages 839–847, 2018. 1

    2018

  5. [5]

    Ruoyu Chen, Jingzhi Li, Hua Zhang, Changchong Sheng, Li Liu, and Xiaochun Cao. Sim2word: Explaining simi- larity with representative attribute words via counterfactual explanations.ACM Transactions on Multimedia Computing, Communications and Applications, 19(6):220:1–22, 2023. 2

    2023

  6. [6]

    Less is more: Fewer interpretable region via submodular subset selection

    Ruoyu Chen, Hua Zhang, Siyuan Liang, Jingzhi Li, and Xi- aochun Cao. Less is more: Fewer interpretable region via submodular subset selection. In12th International Confer- ence on Learning Representations, pages 1–13, 2024. 2

    2024

  7. [7]

    Interpreting object-level foundation models via visual pre- cision search

    Ruoyu Chen, Siyuan Liang, Jingzhi Li, Shiming Liu, Maosen Li, Zhen Huang, Hua Zhang, and Xiaochun Cao. Interpreting object-level foundation models via visual pre- cision search. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 30042–30052, 2025. 2

    2025

  8. [8]

    Less is more: Efficient black-box attribution via minimal interpretable subset selec- tion, 2025

    Ruoyu Chen, Siyuan Liang, Jingzhi Li, Shiming Liu, Li Liu, Hua Zhang, and Xiaochun Cao. Less is more: Efficient black-box attribution via minimal interpretable subset selec- tion, 2025. 1, 2, 3, 8

    2025

Show all 57 references
  1. [9]

    Ruoyu Chen, Hua Zhang, Jingzhi Li, Li Liu, Zhen Huang, and Xiaochun Cao. Generalized semantic contrastive learn- ing via embedding side information for few-shot object de- tection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(8):6496–6514, 2025. 1, 2, 6, 7, 8

    2025

  2. [10]

    Yolo-world: Real-time open-vocabulary object detection

    Tianheng Cheng, Lin Song, Yixiao Ge, Wenyu Liu, Xing- gang Wang, and Ying Shan. Yolo-world: Real-time open-vocabulary object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16901–16911, 2024. 1

    2024

  3. [11]

    Self-interpretable graph learn- ing with sufficient and necessary explanations

    Jiale Deng and Yanyan Shen. Self-interpretable graph learn- ing with sufficient and necessary explanations. InProceed- ings of the AAAI Conference on Artificial Intelligence, pages 11749–11756, 2024. 3

    2024

  4. [12]

    Explanations based on the missing: Towards contrastive ex- planations with pertinent negatives

    Amit Dhurandhar, Pin-Yu Chen, Ronny Luss, Chun-Chen Tu, Paishun Ting, Karthikeyan Shanmugam, and Payel Das. Explanations based on the missing: Towards contrastive ex- planations with pertinent negatives. pages 592–603, 2018. 2

    2018

  5. [13]

    Aga: Attribute-guided augmentation

    Mandar Dixit, Roland Kwitt, Marc Niethammer, and Nuno Vasconcelos. Aga: Attribute-guided augmentation. InPro- ceedings of the IEEE conference on computer vision and pat- tern recognition, pages 7455–7463, 2017. 3

    2017

  6. [14]

    An image is worth 16x16 words: Transformers for image recognition at scale, 2021

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  7. [15]

    Interpretable explana- tions of black boxes by meaningful perturbation

    Ruth C Fong and Andrea Vedaldi. Interpretable explana- tions of black boxes by meaningful perturbation. InPro- ceedings of the IEEE international conference on computer vision, pages 3429–3437, 2017. 2

    2017

  8. [16]

    Going beyond xai: A system- atic survey for explanation-guided learning.ACM Comput

    Yuyang Gao, Siyi Gu, Junji Jiang, Sungsoo Ray Hong, Dazhou Yu, and Liang Zhao. Going beyond xai: A system- atic survey for explanation-guided learning.ACM Comput. Surv., 56(7), 2024. 2

    2024

  9. [17]

    Shortcut learning in deep neural networks

    Robert Geirhos, J ¨orn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Fe- lix A Wichmann. Shortcut learning in deep neural networks. InNature Machine Intelligence, pages 665–673. Springer Nature, 2020. 1

    2020

  10. [18]

    Inter- pretation of neural networks is fragile

    Amirata Ghorbani, Abubakar Abid, and James Zou. Inter- pretation of neural networks is fragile. InProceedings of the AAAI conference on artificial intelligence, pages 3681–3688,

  11. [19]

    Precof: counterfactual explanations for fairness.Machine Learning, 113(5):3111–3142, 2024

    Sofie Goethals, David Martens, and Toon Calders. Precof: counterfactual explanations for fairness.Machine Learning, 113(5):3111–3142, 2024. 2

    2024

  12. [20]

    Lin Guan, Mudit Verma, Suna Sihang Guo, Ruohan Zhang, and Subbarao Kambhampati. Widening the pipeline in human-guided reinforcement learning with explanation and context-aware data augmentation.Advances in Neural Infor- mation Processing Systems, pages 21885–21897, 2021. 3 9

    2021

  13. [21]

    Jianing Hao, Qing Shi, Yilin Ye, and Wei Zeng. Time- tuner: Diagnosing time representations for time-series fore- casting with counterfactual explanations.IEEE Transactions on Visualization and Computer Graphics, 30(1):1183–1193,

  14. [22]

    Deep residual learning for image recognition, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition, 2015. 7

    2015

  15. [23]

    The many faces of robust- ness: A critical analysis of out-of-distribution generalization

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kada- vath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, et al. The many faces of robust- ness: A critical analysis of out-of-distribution generalization. InProceedings of the IEEE/CVF Internationa...

    2021

  16. [24]

    Mambalrp: Explaining selective state space sequence models

    Farnoush Rezaei Jafari, Gr ´egoire Montavon, Klaus-Robert M¨uller, and Oliver Eberle. Mambalrp: Explaining selective state space sequence models. InAdvances in Neural Infor- mation Processing Systems 37 (NeurIPS 2024), pages 1–14,

    2024

  17. [25]

    Drawing attribu- tions from evolved counterfactuals

    Jan Jakubik and Halina Kwa ´snicka. Drawing attribu- tions from evolved counterfactuals. InProceedings of the 2024 Genetic and Evolutionary Computation Conference Companion (GECCO 2024 Companion), pages 1582–1589,

    2024

  18. [26]

    Xrai: Better attributions through regions

    Andrei Kapishnikov, Tolga Bolukbasi, Fernanda Viegas, and Michael Terry. Xrai: Better attributions through regions. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4947–4956, 2019. 2

    2019

  19. [27]

    Omar F. P. Krauss, Marcelo S. Balbino, and Cristiane N. No- bre. Causal csse: integrating counterfactuals and causality in the explanation of machine learning models.Machine Learn- ing, 114(10):221:1–24, 2025. 2

    2025

  20. [28]

    A trustworthy counterfactual explanation method with latent space smoothing.IEEE Transactions on Image Processing, 33(8):4584–4599, 2024

    Yan Li, Xia Cai, Chunwei Wu, Xiao Lin, and Guitao Cao. A trustworthy counterfactual explanation method with latent space smoothing.IEEE Transactions on Image Processing, 33(8):4584–4599, 2024. 2

    2024

  21. [29]

    Ob- ject detectors in the open environment: Challenges, solu- tions, and outlook.arXiv preprint arXiv:2403.16271, 2024

    Siyuan Liang, Wei Wang, Ruoyu Chen, Aishan Liu, Boxi Wu, Ee-Chien Chang, Xiaochun Cao, and Dacheng Tao. Ob- ject detectors in the open environment: Challenges, solu- tions, and outlook.arXiv preprint arXiv:2403.16271, 2024. 1

    2024 arXiv

  22. [30]

    De- coupled metric network for single-stage few-shot object de- tection.IEEE Transactions on Cybernetics, 53(1):514–525,

    Yue Lu, Xingyu Chen, Zhengxing Wu, and Junzhi Yu. De- coupled metric network for single-stage few-shot object de- tection.IEEE Transactions on Cybernetics, 53(1):514–525,

  23. [31]

    Instruct where the model fails: Generative data augmentation via guided self-contrastive fine-tuning

    Weijian Ma, Ruoxin Chen, Keyue Zhang, Shuang Wu, and Shouhong Ding. Instruct where the model fails: Generative data augmentation via guided self-contrastive fine-tuning. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 5991–5999, 2025. 3

    2025

  24. [32]

    Causal explanation-guided learning for organ allo- cation

    Alessandro Marchese, Jeroen Berrevoets, and Sam Ver- boven. Causal explanation-guided learning for organ allo- cation. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, pages 1–14, 2025. 3

    2025

  25. [33]

    Context-aware commonsense knowledge graph reasoning with path-guided explanations.IEEE Trans- actions on Knowledge and Data Engineering, 36(8):3725– 3738, 2024

    Yudai Pan, Jun Liu, Tianzhe Zhao, Lingling Zhang, and Qianying Wang. Context-aware commonsense knowledge graph reasoning with path-guided explanations.IEEE Trans- actions on Knowledge and Data Engineering, 36(8):3725– 3738, 2024. 2

    2024

  26. [34]

    Learn- ing transferable visual models from natural language super- vision, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision, 2021. 7

    2021

  27. [35]

    Imagenet large scale visual recognition challenge.International journal of computer vision, 115(3):211–252, 2015

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, et al. Imagenet large scale visual recognition challenge.International journal of computer vision, 115(3):211–252, 2015. 6

    2015

  28. [36]

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

    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. InProceedings of the IEEE in- ternational conference on computer vision, pages 618–626,

  29. [37]

    Tiny-imagenet-200, 2017

    STANFORD UNIVERSITY . Tiny-imagenet-200, 2017. [DS]. https://cstr.cn/32010.11.sjtu.scidata.00000011. 6

    2017

  30. [38]

    Fsce: Few-shot object detection via contrastive pro- posal encoding

    Bo Sun, Banghuai Li, Shengcai Cai, Ye Yuan, and Chi Zhang. Fsce: Few-shot object detection via contrastive pro- posal encoding. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7352– 7362, 2021. 1

    2021

  31. [39]

    Con- trastive multiview coding

    Yonglong Tian, Dilip Krishnan, and Phillip Isola. Con- trastive multiview coding. InComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XI 16, pages 776–794. Springer,

    2020

  32. [40]

    Effective data augmentation with diffu- sion models

    Brandon Trabucco, Kyle Doherty, Max A Gurinas, and Rus- lan Salakhutdinov. Effective data augmentation with diffu- sion models. InThe Twelfth International Conference on Learning Representations, pages 1–15, 2024. 3

    2024

  33. [41]

    Nomatterxai: Generating ”no matter what” alterfactual ex- amples for explaining black-box text classification models

    Tuc Van Nguyen, James Michels, Hua Shen, and Thai Le. Nomatterxai: Generating ”no matter what” alterfactual ex- amples for explaining black-box text classification models. InProceedings of the AAAI Conference on Artificial Intelli- gence, pages 24939–24947, 2025. 2

    2025

  34. [42]

    Inter- pretable and fine-grained visual explanations for convolu- tional neural networks

    Jorg Wagner, Jan Mathias Kohler, Tobias Gindele, Leon Het- zel, Jakob Thaddaus Wiedemer, and Sven Behnke. Inter- pretable and fine-grained visual explanations for convolu- tional neural networks. InProceedings of the IEEE/CVF conference on computer vision and pattern recogniti...

    2019

  35. [43]

    Learning robust global representations by penalizing local predictive power

    Haohan Wang, Songwei Ge, Zachary Lipton, and Eric P Xing. Learning robust global representations by penalizing local predictive power. InAdvances in Neural Information Processing Systems, 2019. 6

    2019

  36. [44]

    A generalized explanation framework for visualization of deep learning model predic- tions.IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 9265–9283, 2023

    Pei Wang and Nuno Vasconcelos. A generalized explanation framework for visualization of deep learning model predic- tions.IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 9265–9283, 2023. 1

    2023

  37. [45]

    Rethinking maxi- mum mean discrepancy for visual domain adaptation.IEEE Transactions on Neural Networks and Learning Systems, 34 (1):264–277, 2023

    Wei Wang, Haojie Li, Zhengming Ding, Feiping Nie, Jun- yang Chen, Xiao Dong, and Zhihui Wang. Rethinking maxi- mum mean discrepancy for visual domain adaptation.IEEE Transactions on Neural Networks and Learning Systems, 34 (1):264–277, 2023

    2023

  38. [46]

    Deep label 10 propagation with nuclear norm maximization for visual do- main adaptation.IEEE Transactions on Image Processing, 34(1):1246–1258, 2025

    Wei Wang, Hanyang Li, Cong Wang, Chao Huang, Zheng- ming Ding, Feiping Nie, and Xiaochun Cao. Deep label 10 propagation with nuclear norm maximization for visual do- main adaptation.IEEE Transactions on Image Processing, 34(1):1246–1258, 2025. 1

    2025

  39. [47]

    Reinforced path reasoning for counterfactual ex- plainable recommendation.IEEE Transactions on Knowl- edge and Data Engineering, 36(7):3443–3459, 2024

    Xiangmeng Wang, Qian Li, Dianer Yu, Qing Li, and Guan- dong Xu. Reinforced path reasoning for counterfactual ex- plainable recommendation.IEEE Transactions on Knowl- edge and Data Engineering, 36(7):3443–3459, 2024. 2

    2024

  40. [48]

    Graph segmentation and contrastive enhanced explainer for graph neural networks

    Zhiqiang Wang, Jiayu Guo, Jianqing Liang, Jiye Liang, Shiying Cheng, and Jiarong Zhang. Graph segmentation and contrastive enhanced explainer for graph neural networks. InProceedings of the AAAI Conference on Artificial Intel- ligence, pages 21393–21401, 2025. 3

    2025

  41. [49]

    Explanation-based data augmentation for image classifica- tion.Advances in neural information processing systems, 34:20929–20940, 2021

    Sandareka Wickramanayake, Wynne Hsu, and Mong Li Lee. Explanation-based data augmentation for image classifica- tion.Advances in neural information processing systems, 34:20929–20940, 2021. 3

    2021

  42. [50]

    Few-shot object detection and viewpoint estimation for objects in the wild.IEEE Transactions on Pattern Analysis and Machine, 45(3):3090–3106, 2022

    Yang Xiao, Vincent Lepetit, and Renaud Marlet. Few-shot object detection and viewpoint estimation for objects in the wild.IEEE Transactions on Pattern Analysis and Machine, 45(3):3090–3106, 2022. 1

    2022

  43. [51]

    Masked images are counterfactual samples for robust fine-tuning

    Yao Xiao, Ziyi Tang, Pengxu Wei, Cong Liu, and Liang Lin. Masked images are counterfactual samples for robust fine-tuning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20301– 20310, 2023. 1, 2, 6, 7, 8

    2023

  44. [52]

    Knowledge-augmented visual question answer- ing with natural language explanation.IEEE Transactions on Image Processing, 33(3):2652–2664, 2024

    Jiayuan Xie, Yi Cai, Jiali Chen, Ruohang Xu, Jiexin Wang, and Qing Li. Knowledge-augmented visual question answer- ing with natural language explanation.IEEE Transactions on Image Processing, 33(3):2652–2664, 2024. 3

    2024

  45. [53]

    Spatial sensitive grad-cam++: Im- proved visual explanation for object detectors via weighted combination of gradient map

    Toshinori Yamauchi. Spatial sensitive grad-cam++: Im- proved visual explanation for object detectors via weighted combination of gradient map. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8164–8168, 2024. 1

    2024

  46. [54]

    Detclipv3: To- wards versatile generative open-vocabulary object detection

    Lewei Yao, Renjie Pi, Jianhua Han, Xiaodan Liang, Hang Xu, Wei Zhang, Zhenguo Li, and Dan Xu. Detclipv3: To- wards versatile generative open-vocabulary object detection. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 27391–27401, 2024. 3

    2024

  47. [55]

    On the (in) fidelity and sensitivity of explanations.Advances in neural information processing systems, 30(1):1183–1193, 2019

    Chih-Kuan Yeh, Cheng-Yu Hsieh, Arun Suggala, David I In- ouye, and Pradeep K Ravikumar. On the (in) fidelity and sensitivity of explanations.Advances in neural information processing systems, 30(1):1183–1193, 2019. 2

    2019

  48. [56]

    Octet: Object- aware counterfactual explanations

    Mehdi Zemni, Micka ¨el Chen, ´Eloi Zablocki, H ´edi Ben- Younes, Patrick P ´erez, and Matthieu Cord. Octet: Object- aware counterfactual explanations. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 15062–15071, 2023. 2

    2023

  49. [57]

    Learning deep features for discrimi- native localization

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discrimi- native localization. InComputer Vision and Pattern Recog- nition, 2016. 2 11

    2016

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.