Pith. sign in

REVIEW 2 major objections 5 minor 46 references

FiRe: Fixed-Noise Refinement for Visual Counterfactual Explanations

T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read FiRe reframes visual counterfactual explanation as iterative refinement at one fixed noise level, using Pixel Mean Flow to predict clean images and three controls to keep edits localized, achieving about 3x faster inference and 8x fewer…

desk verdict FiRe is a genuine new formulation for visual counterfactuals with real efficiency gains; the paper deserves peer review but needs error bars, a validation protocol, and code before the headline numbers are fully trustworthy. read the letter →

arxiv 2608.08664 v1 pith:HKX25TYI submitted 2026-08-09 cs.CV

Yan Zeng , Changlu Guo , Oskar Kristoffersen , Anders Nymark Christensen , Morten Rieger Hannemose , Anders Bjorholm Dahl This is my paper · ORCID
classification cs.CV
keywords visualcounterfactualexplanationsfixed-noiserefinementPixelMeanFlowdiffusionmodelsclassifierguidancespatialmasksmedicalimagingexplainableAI
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that classifier-guided visual counterfactual explanation does not need to follow a long reverse denoising trajectory. Instead, it maps the input image to a single intermediate noise level and refines that noisy state in place, using Pixel Mean Flow to predict a clean image for the classifier at every step. The authors argue this fixed-noise formulation keeps enough semantic editability while improving spatial locality, and that their three controls—a union hard mask for edit memory, a soft mask for visible changes, and adaptive guidance with early stopping—prevent over-editing and drift. On five tasks across CelebA, CelebA-HQ, and CheXpert, they report about 3x faster online inference and 8x fewer FLOPs than the strongest recent baseline, with comparable or better counterfactual quality. The point would matter because it suggests efficient, localized counterfactual explanations can be built without recursive denoising or low-quality one-step estimates.

What carries the argument

The central mechanism is Pixel Mean Flow used as a one-step clean-image predictor that turns any noisy state at a fixed noise level $t$ into a clean image the classifier can evaluate. Around that predictor, FiRe builds an update loop: a SmoothGrad attribution map selects the top-$\rho$ salient region, a union hard mask accumulates edit regions across steps, a soft mask feathers the visible edit boundary, and an adaptive guidance scale $\eta^k$ strengthens updates when the target probability is far from the flip threshold and weakens them as it approaches. The update is applied to the noisy state under the hard mask, PMF predicts a clean candidate, and early stopping returns the first clean image whose target probability reaches $P_{\mathrm{flip}}$.

What would settle it

A decisive experiment would replace PMF with a Tweedie one-step estimate, keeping everything else fixed; the outcome separates whether the gain comes from fixed-noise refinement or from the one-step predictor.

Watch

Extended reading notes

Core claim

FiRe's central claim is that fixed-noise refinement is a better formulation than DDPM reverse-trajectory editing for visual counterfactual explanation. The paper shows that after forward-noising an input to a fixed level $t=0.4$, iteratively updating the noisy state with classifier gradients backpropagated through a Pixel Mean Flow clean-image predictor produces edits that flip the target classifier with high validity. The method couples this update with a union hard mask that preserves previously selected edit regions, a feathered soft mask that controls visible blending, and a confidence-progress guidance scale with early stopping. The reported experiments support that this formulation reaches the target decision with fewer unintended changes and lower FID than trajectory-based baselines, while cutting compute drastically.

Load-bearing premise

The load-bearing premise is that gradients of the target-classifier loss, sent backward through PMF's one-step clean-image prediction and the soft-mask blend, stay reliable as a descent direction in the fixed-noise state for up to 15 accumulated updates.

Editorial extensions

If this is right

  • Counterfactual generation can run at a single noise level without recursive denoising or low-quality one-step estimates, cutting online inference time roughly threefold and FLOPs roughly eightfold versus MaskDiME.
  • The union hard mask lets previously selected edit regions stay active across refinement steps, so repeated updates at the same noise level accumulate rather than interrupt the counterfactual change.
  • Early stopping at a target-probability threshold trades a modest drop in classifier-output transition for better realism and sparsity; the ablation shows FID rises from 1.98 to 3.47 when early stopping is disabled.
  • The same fixed-noise recipe transfers to a 512x512 medical shortcut-removal task with a different data distribution, indicating the formulation is not face-specific.

Reading between the lines

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

  • A direct test of how much of the gain is due to fixed-noise refinement versus the one-step predictor: replace Pixel Mean Flow with a Tweedie one-step estimate at the same noise level; if flip rate collapses, the gradient path through PMF is the load-bearing part, not the fixed noise level.
  • The union hard mask acts as persistent edit memory, which could be reused for interactive editing or multi-round counterfactual refinement where earlier edits must survive later updates.
  • Applying the same fixed-noise loop inside a latent diffusion or rectified-flow model, where one-step clean-image prediction is already available, is a natural next step; the mask-memory and adaptive-guidance controls are defined for pixel space but should transfer to latent coordinates.
  • Because guidance is computed on a clean image but applied through a learned predictor, the composite gradient may be biased; a calibration experiment comparing PMF-predicted clean-image gradients with ground-truth clean-image gradients at $t=0.4$ would test that bias directly.
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

2 major / 5 minor

Summary. The paper proposes FiRe, a fixed-noise refinement framework for visual counterfactual explanations. Instead of following a DDPM reverse denoising trajectory with changing noise levels, FiRe maps the input to a fixed intermediate noise level and iteratively refines the noisy state at that level. A Pixel Mean Flow (PMF) predictor produces clean classifier-facing images in one step, and the clean-image classifier loss is backpropagated through the PMF and soft-mask blending path to obtain noisy-state gradients. Three controls are introduced for fixed-noise refinement: a dynamic dual-mask strategy (a union hard mask for noisy-state edit memory and a feathered soft mask for clean-image visibility), adaptive guidance that strengthens or weakens updates based on target confidence, and early stopping once the target probability reaches a threshold. The method is evaluated on five tasks across CelebA, CelebA-HQ, and CheXpert, reporting about 3x faster online inference and 8x fewer FLOPs than the strongest recent baseline (MaskDiME) while achieving comparable or better counterfactual quality on most metrics.

Significance. If the claims hold, FiRe is a meaningful step for efficient classifier-guided counterfactual editing: it avoids both recursive denoising and low-quality one-step estimates by combining a fixed-noise optimization state with a one-step clean-image predictor, and it is the first to formulate counterfactual generation as fixed-noise refinement rather than trajectory editing. The paper is transparent: it reruns all baselines under its own CheXpert protocol, reports the full ablation progression from a basic PMF-guided generator to the complete FiRe model, and explicitly discusses the lower COUT values as a design choice. The empirical evidence is substantial, with consistent improvements in FID/sFID and MNAC across four face tasks and a strong medical shortcut-removal result, alongside a clear efficiency comparison.

major comments (2)
  1. [Algorithm 1 / §3.5, Eq. (8)] The computation graph at k=0 is underspecified. Algorithm 1 sets X^0 to be the input image and then, at k=0, computes ∇_{Z_t^0} L_joint^0 via Eq. (8). If X^0 is treated as a constant, then ∂X^0/∂Z_t^0 = 0 and the first update in Eq. (10) is vacuous, meaning the effective refinement budget is K-1 guided steps preceded by a mask-constrained one-step PMF prediction. Alternatively, the authors may intend to differentiate through the inverse of Eq. (2), X^0 = (Z_t^0 - t Z^0)/(1-t), which gives ∂X^0/∂Z_t^0 = I/(1-t) and a nontrivial first step. The text says that ∂X^k/∂Z_t^k follows the path 'including PMF prediction and soft-mask clean-image update', which does not apply to X^0. This ambiguity affects the meaning of the K=15 budget, the reported efficiency accounting, and reproducibility; please specify the intended graph explicitly and, if necessary, correct Algorithm 1.
  2. [§3.5, Eq. (8) / Table 4] The central mechanism of FiRe is the gradient of the clean-image loss transported through the PMF predictor and soft-mask blending back to the fixed-noise state. The paper provides no direct evidence that this composite Jacobian is a reliable descent direction at t=0.4. The ablations show that the full system works and that PMF-Fixed outperforms PMF-Reverse, but they do not isolate the transported gradient from the dual-mask and early-stopping controls; the mask makes a large contribution (FID drops from 19.13 to 3.97 when the FiRe mask is added on top of PMF-Fixed). Please add a diagnostic that directly tests the gradient path, for example by measuring the alignment between ∇_{Z_t} L_joint and the finite-difference change in p_target after a small masked update, or by comparing the fixed-noise refinement with a variant that uses a one-step DDPM Tweedie estimate instead of PMF while keeping all controls identical. This would substantiate the claim that fixed-noise refinement, rather than the spatial controls alone, is responsible for the quality and efficiency gains.
minor comments (5)
  1. [§3.5, Eq. (9)] The notation [·]_+ in Eq. (9) is not defined; please state that it clamps negative values to zero, i.e., [x]_+ = max(x, 0).
  2. [Table 3] The metric MAD is reported but never defined in the text; please define it explicitly in §4.1.
  3. [Table 4] The PMF-Reverse row is described as '15-step refinement along a DDPM-based reverse denoising trajectory', but the noise schedule is not specified. Please provide the schedule or a reference so that the comparison between reverse and fixed refinement is reproducible.
  4. [Figure 3] The marker-size legend for peak GPU memory is described in the caption for panel (a) but panel (b) also uses marker sizes; please clarify whether the same legend applies to both panels.
  5. [Tables 1 and 2] Several entries for DiVE, DiVE100, and STEEX are shown as '–'; please state in the captions whether these values were not reported in the original papers or were not computed in this work.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: FiRe's fixed-noise refinement is an empirical construction grounded in an external one-step predictor and external baselines; the MaskDiME self-citation is comparative, not load-bearing.

full rationale

FiRe's derivation chain is self-contained in the relevant sense. The clean-image predictor, Pixel Mean Flow, is an external generative model cited from prior work by other authors, and the target classifier is fixed and pretrained outside FiRe. The method's equations (2)-(12) define an optimization procedure for producing counterfactuals; none of them sets a predicted quantity equal to its own input by construction. Equation (8) backpropagates clean-image loss gradients to the fixed noisy state, and while the k=0 computation path is under-specified because X^0 is initialized as the original image rather than a PMF output, making ∂X^0/∂Z_t^0 zero, this is an implementation ambiguity or correctness issue rather than a circular reduction. The self-citation to MaskDiME, which shares authors with FiRe, is used as an empirical baseline and as an ablation variant, not as justification for FiRe's central premise; the paper's central claims are evaluated against multiple external methods (DiME, ACE, FastDiME, RCSB, LDCE-txt, TiME) using standard metrics. No fitted parameter is relabeled as a prediction, no uniqueness theorem is imported from the authors, no ansatz is smuggled in via self-citation, and no known result is merely renamed. The fixed-noise refinement, dual-mask strategy, adaptive guidance, and early stopping are new algorithmic components whose quantitative evaluation is external and falsifiable. Therefore, the paper's contribution does not reduce to its inputs. The score is 0 rather than 2 because the only self-citation is to a baseline, not to load-bearing support for the method's validity or uniqueness.

Assumptions & free parameters 10 free parameters · 5 assumptions · 0 invented entities

The method's machinery consists of algorithms built on an external generative model (PMF) and classifier gradients; no new physical or conceptual entities are introduced. The main ledger entries are the hyperparameters chosen on the target tasks and the domain assumptions about PMF's suitability as a differentiable clean-image interface.

free parameters (10)
  • Fixed noise level t = 0.4
    Default in Eq. (2); Table 6 shows t=0.4 is the manually selected best trade-off among {0.2, 0.4, 0.6, 0.8} on CelebA-HQ Smile.
  • Refinement budget K = 15
    Eq. (9) and Algorithm 1; Table 5 selects K from {5,10,15,20} based on FID, sFID, and FR. Over 99% of successful counterfactuals stop within 10 steps.
  • Early stop threshold P_flip = 0.85
    Algorithm 1 stopping condition; chosen by hand as the target probability that counts as a valid counterfactual.
  • Base guidance scale eta = 0.02
    Eq. (9) default base scale; no sensitivity table is reported for this value.
  • Classification loss weight lambda_cls = 1.0
    Eq. (7) loss weight; set by hand, no sensitivity analysis reported.
  • Perceptual loss weight lambda_perc = 0.20
    Eq. (7) loss weight; set by hand, no sensitivity analysis reported.
  • TV loss weight lambda_tv = 0.01
    Eq. (7) loss weight; set by hand, no sensitivity analysis reported.
  • Saliency ratio rho = 5% (CelebA Smile), 12% (CelebA Age), 20% (CheXpert Pacemaker)
    Eq. (4) top-rho salient pixels; task-specific and hand-tuned, no sensitivity table reported for rho.
  • Dilation radius r and feathering width w = r=2, w=3
    Eqs. (4) and (6); mask post-processing parameters chosen by hand, no sensitivity analysis reported.
  • SmoothGrad copies N = 20
    SmoothGrad averaging count in Section 3.4; chosen by hand, no sensitivity analysis reported.
assumptions (5)
  • domain assumption Pixel Mean Flow provides an accurate one-step clean-image predictor G_theta for the noise schedule Z_t=(1-t)X0+tZ0.
    Used in Eq. (11) to produce classifier-facing clean images; taken from [34] without re-derivation or independent verification in this paper.
  • domain assumption Backpropagating the clean-image classifier loss through G_theta and the soft-mask blend yields a useful descent direction for the fixed-noise state.
    Eqs. (8)-(10) define the update; if this composite gradient is biased or unstable, the refinement cannot cross the decision boundary.
  • domain assumption SmoothGrad attribution of target-class logits localizes the pixels whose modification changes the decision.
    Eqs. (3)-(4) use this heat map to build masks; this is a heuristic assumption about the classifier's gradient field.
  • domain assumption Blending the PMF prediction with the original image outside the soft mask keeps decision-irrelevant content intact and yields valid classifier inputs.
    Eq. (12) defines the counterfactual; this assumes the blend does not introduce artifacts that fool the classifier or distort metrics.
  • standard math Softmax(f(X)) defines a meaningful target-class probability for early stopping.
    Eq. (1) defines p_target; standard, but the early-stopping behavior depends on the classifier output being a usable confidence measure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FiRe: Fixed-Noise Refinement for Visual Counterfactual Explanations." pith.science (2026). https://pith.science/paper/HKX25TYI

@misc{pith2026260808664,
  author       = {Pith},
  title        = {Pith review of: FiRe: Fixed-Noise Refinement for Visual Counterfactual Explanations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HKX25TYI}},
  note         = {Machine review of arXiv:2608.08664}
}
abstract

Visual counterfactual explanations aim to change classifier decisions through realistic and localized edits while preserving decision-irrelevant content. Existing DDPM-based methods typically perform classifier-guided editing along a long reverse denoising trajectory. The changing noise levels make semantic editability and spatial control difficult to balance, and the editable state is noisy, whereas the target classifier is trained on clean images. As a result, these methods require either costly recursive denoising or low-quality one-step estimates to obtain classifier-facing clean images. We propose FiRe, a Fixed-noise Refinement framework for visual counterfactual explanations. Rather than following a reverse denoising trajectory, FiRe maps the input to a fixed noise level and iteratively refines the noisy state at that level. To provide clean images for classifier guidance, FiRe first adapts Pixel Mean Flow to visual counterfactual explanation, enabling direct clean-image prediction from noisy states. To make fixed-noise refinement produce minimal and localized counterfactual edits, FiRe introduces three FiRe-specific controls: a dynamic dual-mask strategy, adaptive guidance, and early stopping, which determine where edits accumulate, which changes become visible, and when refinement stops. Experiments on five tasks across three datasets show that, compared with the strongest recent baseline, FiRe achieves about 3$\times$ faster online inference and 8$\times$ fewer FLOPs while obtaining comparable or state-of-the-art counterfactual quality.

Figures

Figures reproduced from arXiv: 2608.08664 by the authors.

Figure 1
Figure 1. Compared with DiME and MaskDiME, FiRe produces a cleaner and more localized edit that better matches the classifier-relevant region. most DDPM-based methods [2, 11, 19, 44] perform classifier-guided editing along a reverse denoising trajectory [15], so counterfactual optimization must proceed through a sequence of changing noise levels. This introduces a timestep-dependent conflict between semantic editability and s… view at source ↗
Figure 2
Figure 2. Overview of FiRe. Given an input image X0, FiRe first applies forward noising to obtain a fixed noisy state Z 0 t , then performs iterative refinement at the same noise level. At step k, the current clean image Xk is used to compute a classifier-driven original mask Mk orig and adaptive gradient guidance ηk∇Z k t L k joint. FiRe derives two masks from Mk orig: a union hard mask Mk hard, which keeps previously select… view at source ↗
Figure 3
Figure 3. Computational cost and generation quality comparison for generating 1,000 counterfactual explanations on the CelebA Smile task (Batch size = 5). (a) Practical infer￾ence time vs. FID, with marker size indicating peak GPU memory. (b) Theoretical compute comparison (Total Peta Floating-Point Operations). FiRe achieves the lowest FID while re￾quiring only 6.37 min and 1.28 PFLOPs, yielding 3× faster inference and 8× fe… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison. FiRe makes the target change while preserving non￾target facial regions and surrounding anatomical structure better than MaskDiME. along a DDPM-based reverse denoising trajectory and substantially improves the flip rate. However, the difference …
Figure 5
Figure 5. Figure 5: Qualitative ablation on the CelebA-HQ Smile task. PMF-One Step produces an insufficient target edit. PMF-Reverse strengthens the smile edit but introduces broad modifications across the face and background, while PMF-Fixed reduces part of this drift but remains spatial…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 34 canonical work pages

  1. [1]

    Explainable artificial intelli- gence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI.Information Fusion, 58:82–115, 2020

    Alejandro Barredo Arrieta, Natalia Díaz-Rodríguez, Javier Del Ser, Adrien Bennetot, Siham Tabik, Alberto Barbado, Salvador García, Sergio Gil-López, Daniel Molina, Richard Benjamins, Raja Chatila, and Francisco Herrera. Explainable artificial intelli- gence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI.Information Fusion,...

    work page 2020

  2. [2]

    Diffu- sion visual counterfactual explanations.Advances in Neural Information Processing Systems, 35:364–377, 2022

    Maximilian Augustin, Valentyn Boreiko, Francesco Croce, and Matthias Hein. Diffu- sion visual counterfactual explanations.Advances in Neural Information Processing Systems, 35:364–377, 2022

    work page 2022

  3. [3]

    Sparse visual counterfactual explanations in image space

    Valentyn Boreiko, Maximilian Augustin, Francesco Croce, Philipp Berens, and Matthias Hein. Sparse visual counterfactual explanations in image space. InGerman Conference on Pattern Recognition, pages 133–148, 2022

    work page 2022

  4. [4]

    Parkhi, and Andrew Zisserman

    Qiong Cao, Li Shen, Weidi Xie, Omkar M. Parkhi, and Andrew Zisserman. VGGFace2: A dataset for recognising faces across pose and age. InProceedings of the IEEE Inter- national Conference on Automatic Face and Gesture Recognition, pages 67–74, 2018

    work page 2018

  5. [5]

    Exploring simple Siamese representation learning

    Xinlei Chen and Kaiming He. Exploring simple Siamese representation learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pages 15750–15758, 2021

    work page 2021

  6. [6]

    StarGAN: Unified generative adversarial networks for multi-domain image-to- image translation

    Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and Jaegul Choo. StarGAN: Unified generative adversarial networks for multi-domain image-to- image translation. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 8789–8797, 2018

    work page 2018

  7. [7]

    Diffusion models beat GANs on image syn- thesis.Advances in Neural Information Processing Systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat GANs on image syn- thesis.Advances in Neural Information Processing Systems, 34:8780–8794, 2021. 16ZENG ET AL.: FIRE

    work page 2021

  8. [8]

    Latent diffusion counter- factual explanations.arXiv preprint arXiv:2310.06668, 2023

    Karim Farid, Simon Schrodi, Max Argus, and Thomas Brox. Latent diffusion counter- factual explanations.arXiv preprint arXiv:2310.06668, 2023

    arXiv 2023

Show all 46 references
  1. [9]

    DeCoDEx: Confounder detector guidance for improved diffusion-based counterfactual explanations.arXiv preprint arXiv:2405.09288, 2024

    Nima Fathi, Amar Kumar, Brennan Nichyporuk, Mohammad Havaei, and Tal Arbel. DeCoDEx: Confounder detector guidance for improved diffusion-based counterfactual explanations.arXiv preprint arXiv:2405.09288, 2024

    2024 arXiv

  2. [10]

    Counter- factual visual explanations

    Yash Goyal, Ziyan Wu, Jan Ernst, Dhruv Batra, Devi Parikh, and Stefan Lee. Counter- factual visual explanations. InInternational Conference on Machine Learning, pages 2376–2384, 2019

    2019

  3. [11]

    MaskDiME: Adaptive masked diffusion for precise and efficient visual counterfactual explanations.arXiv preprint arXiv:2602.18792, 2026

    Changlu Guo, Anders Nymark Christensen, Anders Bjorholm Dahl, and Morten Rieger Hannemose. MaskDiME: Adaptive masked diffusion for precise and efficient visual counterfactual explanations.arXiv preprint arXiv:2602.18792, 2026

    2026 arXiv

  4. [12]

    AttGAN: Facial attribute editing by only changing what you want.IEEE Transactions on Image Processing, 28(11):5464–5478, 2019

    Zhenliang He, Wangmeng Zuo, Meina Kan, Shiguang Shan, and Xilin Chen. AttGAN: Facial attribute editing by only changing what you want.IEEE Transactions on Image Processing, 28(11):5464–5478, 2019

    2019

  5. [13]

    Generating visual explanations

    Lisa Anne Hendricks, Zeynep Akata, Marcus Rohrbach, Jeff Donahue, Bernt Schiele, and Trevor Darrell. Generating visual explanations. InEuropean Conference on Com- puter Vision, pages 3–19, 2016

    2016

  6. [14]

    GANs trained by a two time-scale update rule converge to a local Nash equilibrium.Advances in Neural Information Processing Systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. GANs trained by a two time-scale update rule converge to a local Nash equilibrium.Advances in Neural Information Processing Systems, 30, 2017

    2017

  7. [15]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    2020

  8. [16]

    Weinberger

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q. Weinberger. Densely connected convolutional networks. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition, pages 4700–4708, 2017

    2017

  9. [17]

    CheXpert: A large chest radiograph dataset with uncertainty labels and expert com- parison

    Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, et al. CheXpert: A large chest radiograph dataset with uncertainty labels and expert com- parison. InProceedings of the AAAI...

    2019

  10. [18]

    STEEX: Steering counterfactual explanations with semantics

    Paul Jacob, Éloi Zablocki, Hedi Ben-Younes, Mickaël Chen, Patrick Pérez, and Matthieu Cord. STEEX: Steering counterfactual explanations with semantics. InEu- ropean Conference on Computer Vision, pages 387–403, 2022

    2022

  11. [19]

    Diffusion models for counterfac- tual explanations

    Guillaume Jeanneret, Loïc Simon, and Frédéric Jurie. Diffusion models for counterfac- tual explanations. InProceedings of the Asian Conference on Computer Vision, pages 858–876, 2022

    2022

  12. [20]

    Adversarial counterfactual visual explanations

    Guillaume Jeanneret, Loïc Simon, and Frédéric Jurie. Adversarial counterfactual visual explanations. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16425–16435, 2023. ZENG ET AL.: FIRE17

    2023

  13. [21]

    Text-to-image models for coun- terfactual explanations: A black-box approach

    Guillaume Jeanneret, Loïc Simon, and Frédéric Jurie. Text-to-image models for coun- terfactual explanations: A black-box approach. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 4757–4767, 2024

    2024

  14. [22]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. InEuropean Conference on Computer Vision, pages 694–711, 2016

    2016

  15. [23]

    xGEMs: Gener- ating exemplars to explain black-box models.arXiv preprint arXiv:1806.08867, 2018

    Shalmali Joshi, Oluwasanmi Koyejo, Been Kim, and Joydeep Ghosh. xGEMs: Gener- ating exemplars to explain black-box models.arXiv preprint arXiv:1806.08867, 2018

    2018 arXiv

  16. [24]

    Progressive growing of GANs for improved quality, stability, and variation.arXiv preprint arXiv:1710.10196, 2017

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of GANs for improved quality, stability, and variation.arXiv preprint arXiv:1710.10196, 2017

    2017 arXiv

  17. [25]

    Cycle-consistent counterfactuals by latent transforma- tions

    Saeed Khorram and Li Fuxin. Cycle-consistent counterfactuals by latent transforma- tions. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10203–10212, 2022

    2022

  18. [26]

    Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (TCA V)

    Been Kim, Martin Wattenberg, Justin Gilmer, Carrie Cai, James Wexler, Fernanda Vié- gas, et al. Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (TCA V). InInternational Conference on Machine Learning, pages 2668–2677, 2018

    2018

  19. [27]

    PRISM: High-resolution and precise counterfactual medical image generation using language-guided stable dif- fusion.arXiv preprint arXiv:2503.00196, 2025

    Amar Kumar, Anita Kriz, Mohammad Havaei, and Tal Arbel. PRISM: High-resolution and precise counterfactual medical image generation using language-guided stable dif- fusion.arXiv preprint arXiv:2503.00196, 2025

    2025 arXiv

  20. [28]

    Freeman, Phillip Isola, Amir Globerson, Michal Irani, et al

    Oran Lang, Yossi Gandelsman, Michal Yarom, Yoav Wald, Gal Elidan, Avinatan Has- sidim, William T. Freeman, Phillip Isola, Amir Globerson, Michal Irani, et al. Explain- ing in style: Training a GAN to explain a classifier in StyleSpace. InProceedings of the IEEE/CVF Internation...

    2021

  21. [29]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    2022 arXiv

  22. [30]

    Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022

    Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022

    2022 arXiv

  23. [31]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InInternational Conference on Learning Representations, 2023

    2023

  24. [32]

    InstaFlow: One step is enough for high-quality diffusion-based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, and Qiang Liu. InstaFlow: One step is enough for high-quality diffusion-based text-to-image generation. InInterna- tional Conference on Learning Representations, 2024

    2024

  25. [33]

    Deep learning face attributes in the wild

    Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. InProceedings of the IEEE International Conference on Computer Vision, pages 3730–3738, 2015. 18ZENG ET AL.: FIRE

    2015

  26. [34]

    One-step latent-free image generation with Pixel Mean Flows.arXiv preprint arXiv:2601.22158, 2026

    Yiyang Lu, Susie Lu, Qiao Sun, Hanhong Zhao, Zhicheng Jiang, Xianbang Wang, Tianhong Li, Zhengyang Geng, and Kaiming He. One-step latent-free image generation with Pixel Mean Flows.arXiv preprint arXiv:2601.22158, 2026

    2026 arXiv

  27. [35]

    Herbert E. Robbins. An empirical Bayes approach to statistics. InBreakthroughs in Statistics: Foundations and Basic Theory, pages 388–394. Springer, 1992

    1992

  28. [36]

    Beyond trivial counterfactual explanations with diverse valuable explanations

    Pau Rodriguez, Massimo Caccia, Alexandre Lacoste, Lee Zamparo, Issam Laradji, Laurent Charlin, and David Vazquez. Beyond trivial counterfactual explanations with diverse valuable explanations. InProceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 1...

    2021

  29. [37]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Om- mer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684– 10695, 2022

    2022

  30. [38]

    Rudin, Stanley Osher, and Emad Fatemi

    Leonid I. Rudin, Stanley Osher, and Emad Fatemi. Nonlinear total variation based noise removal algorithms.Physica D: Nonlinear Phenomena, 60(1–4):259–268, 1992

    1992

  31. [39]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    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 International Conference on Computer Vision, pages 618–626, 2017

    2017

  32. [40]

    Deep inside convolutional networks: Visualising image classification models and saliency maps.arXiv preprint arXiv:1312.6034, 2013

    Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps.arXiv preprint arXiv:1312.6034, 2013

    2013 arXiv

  33. [41]

    SmoothGrad: Removing noise by adding noise.arXiv preprint arXiv:1706.03825, 2017

    Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Viégas, and Martin Wattenberg. SmoothGrad: Removing noise by adding noise.arXiv preprint arXiv:1706.03825, 2017

    2017 arXiv

  34. [42]

    Rethinking visual counterfactual explanations through region con- straint

    Bartlomiej Sobieski, Jakub Grzywaczewski, Bartłomiej Sadlej, Matthew Tivnan, and Przemyslaw Biecek. Rethinking visual counterfactual explanations through region con- straint. InInternational Conference on Learning Representations, 2025

    2025

  35. [43]

    Counterfactual explanations without opening the black box: Automated decisions and the GDPR.Harvard Journal of Law & Technology, 31(2):841–887, 2017

    Sandra Wachter, Brent Mittelstadt, and Chris Russell. Counterfactual explanations without opening the black box: Automated decisions and the GDPR.Harvard Journal of Law & Technology, 31(2):841–887, 2017

    2017

  36. [44]

    Fast diffusion-based counterfactuals for shortcut removal and generation

    Nina Weng, Paraskevas Pegios, Eike Petersen, Aasa Feragen, and Siavash Bigdeli. Fast diffusion-based counterfactuals for shortcut removal and generation. InEuropean Con- ference on Computer Vision, pages 338–357, 2024

    2024

  37. [45]

    Text-to- image rectified flow as plug-and-play priors

    Xiaofeng Yang, Cheng Chen, Xulei Yang, Fayao Liu, and Guosheng Lin. Text-to- image rectified flow as plug-and-play priors. InInternational Conference on Learning Representations, 2025

    2025

  38. [46]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 586–595, 2018

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.