Pith. sign in

REVIEW 2 major objections 2 minor 28 references

LLIFT produces semi-synthetic brain MRIs whose lesions are confined to user-specified masks, using only weak labels, and matches the distributional gap between real healthy and pathological scans.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 14:01 UTC pith:IEGVSVE3

load-bearing objection Useful XAI benchmark construction, but the FID realism claim is structurally blind to the edited mask; worth a serious referee with a demand for patch-level evaluation. the 2 major comments →

arxiv 2607.18882 v1 pith:IEGVSVE3 submitted 2026-07-21 cs.CV cs.LG

Local Label-Informed Feature Transfer for Generating Ground-Truth Medical Images: A Comparison of GAN- and Diffusion-Based Approaches

Rick Wilming , Irem Ozseker , Luca Matteo Cornils , Ahc\`ene Boubekki , Benedict Clark , Danny Panknin , Stefan Haufe This is my paper
classification cs.CV cs.LG
keywords local label-informed feature transfersemi-synthetic brain MRIweakly supervised lesion synthesisGANdiffusion modelimage inpaintingXAI benchmarkingground-truth medical images
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper is trying to establish a practical route to ground-truth medical images for XAI validation: take a real healthy brain MRI, specify a mask, and have a generative model insert a plausible lesion inside that mask while leaving everything outside untouched. The key claim is that this can be done without pixel-level lesion annotations—LLIFT-GAN learns from the binary healthy-versus-pathological label alone, and LLIFT-DM learns from bounding-box masks whose locations are shuffled away from the actual lesions during training. Both variants reach Fréchet Inception Distance scores on the order of the natural inter-class reference, meaning their outputs sit roughly as far from healthy images as real pathological images do. A sympathetic reader would care because this gives attribution methods a benchmark where the informative region is known by construction, avoiding expert annotation noise and unrealistic geometric perturbations.

Core claim

LLIFT's central discovery is that the visual signature of pathology can be transferred into a chosen region of a healthy image under weak supervision, and that two very different generative paradigms can do this equally well. LLIFT-GAN extends DCGAN with a dual-view discriminator and an architectural constraint (Eq. 1) so the generator's output is confined to the mask; the discriminator's masked-out view acts as a prior that forces the generated lesion to be convincing at the global scale. LLIFT-DM fine-tunes a large pretrained latent diffusion model with a ControlNet-style inpainting branch, shuffling ground-truth lesion masks across samples so the model learns the appearance of lesions rat

What carries the argument

The central machinery is the mask-confined generation interface shared by both variants: a healthy slice plus a binary mask are the inputs, and the output is identical to the healthy slice outside the mask (enforced architecturally in Eq. 1 for LLIFT-GAN and by post-hoc blending in Eq. 4 for LLIFT-DM). Weak supervision comes from two mechanisms: LLIFT-GAN's dual-view discriminator, which combines a full-image view with a masked-out view so the critic cannot ignore small lesions and the generator must produce globally convincing pathology; and LLIFT-DM's shuffled-mask conditioning, which decouples lesion location from image content so the diffusion model learns the appearance of lesions rathe

Load-bearing premise

The load-bearing premise is that the smoothed-noise synthetic lesions that define the pathological class are representative enough of real clinical pathology that a model trained and evaluated against them produces genuinely realistic lesions; if that premise fails, the ground-truth claim collapses.

What would settle it

Compute a mask-restricted FID (or train a healthy-versus-pathological patch classifier) on the generated patches only; if generated patches are no closer to real pathological patches than healthy patches are, the claim that LLIFT produces realistic lesions is refuted.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Attribution methods can be benchmarked on images where the informative region is known exactly, without expert pixel-level segmentations.
  • The same framework should transfer to other pathologies and modalities wherever a binary class label or coarse mask exists.
  • The healthy-image-plus-mask-plus-output structure makes each benchmark sample self-verifying: outside the mask, output and input are pixel-identical.
  • The two paradigms offer a clear trade-off: discriminative supervision with unstable training (GAN) versus strong visual priors with stable fine-tuning (diffusion).
  • Reporting FID with intra- and inter-class references gives later methods a common scale for judging whether generated pathology is as distinct from healthy data as real pathology is.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: the headline FID values are dominated by the unchanged healthy background, so equal-to-inter-class FID is a weak test of patch realism; a mask-localized FID or a classifier on the generated patches alone would be the decisive experiment.
  • Editorial inference: because the pathological training signal is itself synthetic (smoothed-noise lesions), 'realistic' in this paper means structurally coherent in the MRI setting, not clinically validated; re-running the framework with real lesion segmentations would settle the clinical question.
  • Editorial inference: the shuffled-mask trick is a general recipe for forcing any inpainting model to learn content rather than location, and is worth testing on non-medical conditional generation tasks.
  • Editorial inference: the paper's own goal suggests a direct validation experiment—run attribution methods on LLIFT images and measure whether the highlighted regions coincide with the masks; that would test the informativeness assumption the benchmark is built for.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper introduces LLIFT, a framework for generating semi-synthetic brain MR images with user-controlled lesion placement, intended as ground-truth data for evaluating XAI attribution methods. Two instantiations are presented: LLIFT-GAN, a custom GAN trained on binary healthy/pathological labels with a mask-restricted generator (Eq. (1)), and LLIFT-DM, a Stable Diffusion/ControlNet pipeline conditioned on bounding-box masks with post-hoc blending (Eq. (4)). Both paradigms guarantee pixel-wise identity outside the mask. The authors report FID scores relative to inter-class and intra-class references (Table I) and claim that both variants achieve FID values 'comparable to the natural inter-class reference,' implying that generated lesions are pathologically realistic. The paper also includes qualitative examples and a discussion of limitations.

Significance. If the central claim were substantiated, the LLIFT framework would fill a genuine gap: providing spatially known, class-informative features for XAI validation without pixel-level lesion annotations. The architectural guarantees in Eq. (1) and Eq. (4), the weak-supervision setup, and the explicit comparison of GAN and diffusion paradigms are valuable contributions. The paper is also commendably transparent in Section VI about the limitations of image-level FID and the synthetic nature of the lesion data. However, these self-acknowledged limitations directly undercut the load-bearing quantitative claim in the abstract and Section VII. The evaluation does not currently demonstrate that the generated lesions are realistic or that the known mask contains class-informative features, which is the stated purpose of the benchmark. With additional task-based or mask-restricted evaluation, the framework could become a solid contribution.

major comments (2)
  1. [§VII and Abstract] The abstract and conclusion overstate the clinical realism. Section III defines the pathological distribution as data from [7] in which 'smoothed-noise lesions with regular and irregular compactness are inserted into otherwise healthy slices.' Thus, both the training signal and every FID reference are built from synthetic noise lesions, not clinical pathology. The introduction claims 'clinically plausible lesion-like content' and the abstract claims 'realistic lesion structures,' but these are not supported by the evaluation. Section VI acknowledges that 'replacing the synthetic lesions with real clinical pathology is the natural next step,' which is a direct admission that the current benchmark does not establish clinical realism. If the learned lesion signature and the ground-truth masks are inherited from a synthetic insertion process, the claim that LLIFT provides ground-truth inform
  2. [§VI and §VII] The paper contains a contradiction between the limitations discussion and the conclusion. Section VI explicitly states that 'The image-level FID is limited as an evaluation metric when only a small region of the image is modified' and that 'Metrics operating at the scale of the edited region... would be more sensitive to the actual content of the lesion patch.' Section VII nonetheless repeats the FID-based claim as a central result. A quantitative evaluation that the authors themselves identify as blind to the edited region cannot serve as the primary evidence for the benchmark's validity. The manuscript needs either a new evaluation at the patch level (e.g., masked FID, a separately trained pathology classifier, or a human reader study) or a significant downscoping of the claims to 'structurally controlled synthetic lesions' with the understanding that informativeness is not yet demonst
minor comments (2)
  1. [§IV-A, Eq. (2)] The discriminator objective near Eq. (2) is described as 'the smoothed adversarial objective: min_θ max_ϕ (1−α)E_{x_u,m}[φ] + E_{x_h,z,m}[ψ]' with φ = log D_ϕ(x_u,m) and ψ = log(1−D_ϕ(G_θ(...),m)). The min/max roles are not fully specified: the generator should minimize the same objective while the discriminator maximizes it. The notation as written suggests the generator minimizes a sum that includes the real-data log-likelihood, which is unusual. Please clarify the exact optimizers and the role of α in the two terms.
  2. [General] There is a typo in §IV: 'Two guiding principles steer LLFIT’s design' should read 'LLIFT.' Also, Table I would benefit from showing the number of samples per FID computation and confidence intervals, as FID estimates on small batches are noisy.

Circularity Check

1 steps flagged

Headline FID success is an artifact of the mask-preserving construction, not a measure of learned lesion realism.

specific steps
  1. self definitional [Section IV-A Eq. (1) and Eq. (4); Section IV-B; Section V-B Table I; Section VII]
    "Gθ(xh, m, z) = (1−m)⊙xh +m⊙ ˜Gθ(xh, m, z) ... the overall image-level FID is necessarily dominated by the unchanged healthy background. ... A generator that reaches the inter-class reference has, by definition, produced a batch that is distributionally as different from healthy data as real pathological data is."

    Every output equals the healthy input outside a small mask (Eq. 1 for LLIFT-GAN, Eq. 4 for LLIFT-DM). FID is computed on full images, so the unchanged background drives the distance. Table I shows the generated-vs-pathological FID is essentially the inter-class reference (41.69 vs 41.75 for LLIFT-GAN; blended 7.61 vs 5.84 for LLIFT-DM, while blended-vs-healthy is 4.78). An identity-on-background generator would reproduce these values regardless of the learned patch content. The headline claim that both variants 'reach FID scores comparable to the inter-class reference' is thus a consequence of the construction, not an empirical validation that the generated lesions are pathological.

full rationale

The paper's load-bearing quantitative claim is that both LLIFT variants reach FID scores comparable to the inter-class reference, which is then used to support the conclusion that they generate realistic pathological content suitable as XAI ground truth. That claim reduces by construction: Eq. (1) and Eq. (4) force every generated image to be pixel-identical to the healthy input outside the user-specified mask, and the FID is a full-image Inception-v3 statistic. With small masks, FID(generated, pathological) is dominated by unchanged healthy tissue and is therefore expected to approximate FID(healthy, pathological), exactly the inter-class reference. Table I confirms this numerically, and the paper itself concedes in Section VI: 'The image-level FID is limited as an evaluation metric when only a small region of the image is modified.' Nevertheless Section VII retains the FID comparison as a headline result. This is a partial circularity: the success metric is defined so that the architecture trivially satisfies it; any mask-preserving generator, including a constant fill, would do so. The additional concern that the 'pathological' distribution is itself the same group's synthetic [7] lesions is a validity/clinical-realism limitation rather than a separate circularity, so it does not further increase the score. No self-citation uniqueness chain or ansatz-smuggling was found. Score 6: one central 'prediction' reduces by construction, but the known-mask property and qualitative inspection give the paper some independent content.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The free parameters are mostly standard training choices; the one that materially shapes the result is the choice of image-level FID as the evaluation lens. The axioms that matter are domain assumptions about the synthetic pathological dataset and the validity of whole-image FID for small edits.

free parameters (5)
  • Label smoothing α (LLIFT-GAN) = not specified (in (0,1))
    One-sided label smoothing on the real branch of the adversarial objective; chosen by hand, affects training stability and result quality.
  • Box expansion padding r (LLIFT-DM) = not specified
    Expansion of the smallest enclosing rectangle into the binary mask; controls the edit region size and is chosen without sensitivity analysis.
  • Number of bounding boxes at inference = 1–4
    Sampled per test image; affects lesion load and FID; no ablation or justification.
  • Training hyperparameters = GAN: lr 1e-3 to 1e-5; DM: lr 1e-5, 8 epochs, 3 cosine cycles
    Learning-rate schedules and epoch counts are chosen without ablations; the paper notes GAN performance is seed-dependent.
  • Fixed text prompt for LLIFT-DM = "A brain MRI with a lesion"
    Chosen as a placeholder for the text-conditioning interface; its effect on generation is not studied.
axioms (4)
  • domain assumption The 'real pathological' distribution consists of smoothed-noise lesions inserted into healthy HCP slices (ref [7])
    Section III defines the dataset; all pathological reference images are themselves synthetic, so clinical realism is not addressed.
  • domain assumption A binary class label or shuffled-mask supervision is sufficient to learn lesion-like visual features
    Core premise of LLIFT; no formal guarantee, only empirical FID and qualitative inspection.
  • domain assumption Image-level Inception-v3 FID is a meaningful measure of lesion generation quality when edits are confined to small masks
    Used in Section V-B; the paper later concedes in the Discussion that image-level FID is limited when only a small region is modified.
  • standard math Standard GAN and latent diffusion training objectives converge as expected
    Sections IV-A and IV-B rely on the standard adversarial loss and flow-matching noise-prediction loss without convergence guarantees beyond empirical practice.

pith-pipeline@v1.3.0-alltime-deepseek · 8757 in / 13232 out tokens · 122115 ms · 2026-08-01T14:01:56.453385+00:00 · methodology

0 comments
read the original abstract

Validating Explainable Artificial Intelligence (XAI) methods in medical imaging requires ground-truth data with known locations of informative features. However, current approaches rely on expert annotations, which are prone to labeling errors, or on hand-crafted artificial perturbations superimposed onto healthy images to mimic lesions or malignant features, which lack clinical realism. We present Local Label-Informed Feature Transfer (LLIFT), a framework for generating semi-synthetic brain magnetic resonance images with realistic lesions placed in user-controlled regions, which does not require pixel-level lesion annotations during training. We implement LLIFT with two generative paradigms: LLIFT-GAN, a custom GAN that learns pathological features from binary class labels alone, and LLIFT-DM, a diffusion-based inpainting pipeline conditioned on bounding-box masks via ControlNet. Both approaches are evaluated on brain magnetic resonance imaging data derived from the Human Connectome Project. In evaluations, both achieve Fr\'echet Inception Distance scores, with respect to the real pathological distribution, that are comparable to the inter-class reference between healthy and pathological images in the given dataset. Furthermore, qualitative inspection confirms the realism of lesion structures. The resulting benchmark datasets provide spatially controlled ground truth data for evaluating XAI methods in medical imaging.

Figures

Figures reproduced from arXiv: 2607.18882 by Ahc\`ene Boubekki, Benedict Clark, Danny Panknin, Irem Ozseker, Luca Matteo Cornils, Rick Wilming, Stefan Haufe.

Figure 1
Figure 1. Figure 1: Examples of generated pathological images using LLIFT-GAN. From [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Examples of generated pathological images with artifacts using LLIFT [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Examples of generated pathological images using LLIFT-DM. From [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 5 linked inside Pith

  1. [1]

    Explainable artificial intelligence (XAI) in deep learning- based medical image analysis,

    B. H. M. van der Velden, H. J. Kuijf, K. G. A. Gilhuijs, and M. A. Viergever, “Explainable artificial intelligence (XAI) in deep learning- based medical image analysis,”Medical Image Analysis, vol. 79, p. 102470, 2022

    2022

  2. [2]

    Theoretical behavior of XAI methods in the presence of suppressor variables,

    R. Wilming, L. Kieslich, B. Clark, and S. Haufe, “Theoretical behavior of XAI methods in the presence of suppressor variables,” inProceedings of the 40th International Conference on Machine Learning(A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, and J. Scarlett, eds.), vol. 202 ofProceedings of Machine Learning Research, pp. 37091– 37107, PMLR, 2023

    2023

  3. [3]

    Feature salience - not task-informativeness - drives machine learning model explanations,

    B. Clark, M. Oliveira, R. Wilming, and S. Haufe, “Feature salience - not task-informativeness - drives machine learning model explanations,” 2026

    2026

  4. [4]

    Scrutinizing XAI using linear ground-truth data with suppressor variables,

    R. Wilming, C. Budding, K.-R. M ¨uller, and S. Haufe, “Scrutinizing XAI using linear ground-truth data with suppressor variables,”Machine Learning, vol. 111, no. 5, pp. 1903–1923, 2022

    1903

  5. [5]

    XAI-TRIS: Non-linear image benchmarks to quantify false positive post-hoc attribution of feature importance,

    B. Clark, R. Wilming, and S. Haufe, “XAI-TRIS: Non-linear image benchmarks to quantify false positive post-hoc attribution of feature importance,”Machine Learning, vol. 113, no. 9, pp. 6871–6910, 2024

    2024

  6. [6]

    Position: XAI needs formal notions of explanation correctness,

    S. Haufe, R. Wilming, B. Clark, R. Zhumagambetov, D. Panknin, and A. Boubekki, “Position: XAI needs formal notions of explanation correctness,” inInterpretable AI: Past, Present and Future, 2024. 5

    2024

  7. [7]

    Benchmarking the influence of pre-training on explanation performance in MR image classification,

    M. Oliveira, R. Wilming, B. Clark, C. Budding, F. Eitel, K. Ritter, and S. Haufe, “Benchmarking the influence of pre-training on explanation performance in MR image classification,”Frontiers in Artificial Intelli- gence, vol. 7, p. 1330919, 2024

    2024

  8. [8]

    FunnyNodules: A Cus- tomizable Medical Dataset Tailored for Evaluating Explainable AI

    L. Gall ´ee, Y . Xiong, M. Beer, and M. G ¨otz, “FunnyNodules: A Cus- tomizable Medical Dataset Tailored for Evaluating Explainable AI.”

  9. [9]

    The brain tumor segmentation (brats) challenge: Local synthesis of healthy brain tissue via inpainting,

    F. Kofler, F. Meissen, F. Steinbauer, R. Graf, S. K. Ehrlich, A. Reinke, E. Oswald, D. Waldmannstetter, F. Hoelzl, I. Horvath,et al., “The brain tumor segmentation (brats) challenge: Local synthesis of healthy brain tissue via inpainting,”arXiv preprint arXiv:2305.08992, 2023

    Pith/arXiv arXiv 2023

  10. [10]

    Fastsurfer- lit: Lesion inpainting tool for whole-brain mri segmentation with tumors, cavities, and abnormalities,

    C. Pollak, D. K ¨ugler, T. Bauer, T. R ¨uber, and M. Reuter, “Fastsurfer- lit: Lesion inpainting tool for whole-brain mri segmentation with tumors, cavities, and abnormalities,”Imaging neuroscience, vol. 3, p. imag a 00446, 2025

    2025

  11. [11]

    Lesion region inpainting: an approach for pseudo-healthy image synthesis in intracra- nial infection imaging,

    X. Liu, C. Xiang, L. Lan, C. Li, H. Xiao, and Z. Liu, “Lesion region inpainting: an approach for pseudo-healthy image synthesis in intracra- nial infection imaging,”Frontiers in Microbiology, vol. 15, p. 1453870, 2024

    2024

  12. [12]

    Denoising diffusion models for inpainting of healthy brain tissue,

    A. Durrer, P. C. Cattin, and J. Wolleb, “Denoising diffusion models for inpainting of healthy brain tissue,” inInternational Challenge on Cross-Modality Domain Adaptation for Medical Image Segmentation, pp. 35–45, Springer, 2023

    2023

  13. [13]

    Medical image synthesis for data augmentation and anonymization using generative adversarial networks,

    H.-C. Shin, N. A. Tenenholtz, J. K. Rogers, C. G. Schwarz, M. L. Senjem, J. L. Gunter, K. P. Andriole, and M. Michalski, “Medical image synthesis for data augmentation and anonymization using generative adversarial networks,” inInternational workshop on simulation and synthesis in medical imaging, pp. 1–11, Springer, 2018

    2018

  14. [14]

    Synthesis of brain tumor multicontrast mr images for improved data augmentation,

    S. Kim, B. Kim, and H. Park, “Synthesis of brain tumor multicontrast mr images for improved data augmentation,”Medical Physics, vol. 48, no. 5, pp. 2185–2198, 2021

    2021

  15. [15]

    Multitask brain tumor inpainting with diffusion models: a methodological report,

    P. Rouzrokh, B. Khosravi, S. Faghani, M. Moassefi, S. Vahdati, and B. J. Erickson, “Multitask brain tumor inpainting with diffusion models: a methodological report,”arXiv preprint arXiv:2210.12113, 2022

    Pith/arXiv arXiv 2022

  16. [16]

    Lefusion: Controllable pathology synthesis via lesion-focused diffusion models,

    H. Zhang, Y . Liu, J. Yang, S. Wan, X. Wang, W. Peng, and P. Fua, “Lefusion: Controllable pathology synthesis via lesion-focused diffusion models,” inInternational Conference on Learning Representations, vol. 2025, pp. 13232–13253, 2025

    2025

  17. [17]

    Explaining classifiers with causal concept effect (cace),

    Y . Goyal, A. Feder, U. Shalit, and B. Kim, “Explaining classifiers with causal concept effect (cace),”arXiv preprint arXiv:1907.07165, 2019

    Pith/arXiv arXiv 1907

  18. [18]

    Unsupervised representation learning with deep convolutional generative adversarial networks,

    A. Radford, L. Metz, and S. Chintala, “Unsupervised representation learning with deep convolutional generative adversarial networks,”arXiv preprint arXiv:1511.06434, 2015

    Pith/arXiv arXiv 2015

  19. [19]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10684–10695, 2022

    2022

  20. [20]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” inProceedings of the IEEE/CVF international conference on computer vision, pp. 3836–3847, 2023

    2023

  21. [21]

    Roentgen: vision-language foundation model for chest x-ray generation,

    P. Chambon, C. Bluethgen, J.-B. Delbrouck, R. Van der Sluijs, M. Połacin, J. M. Z. Chaves, T. M. Abraham, S. Purohit, C. P. Langlotz, and A. Chaudhari, “Roentgen: vision-language foundation model for chest x-ray generation,”arXiv preprint arXiv:2211.12737, 2022

    Pith/arXiv arXiv 2022

  22. [22]

    Repaint: Inpainting using denoising diffusion probabilistic models,

    A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool, “Repaint: Inpainting using denoising diffusion probabilistic models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 11461–11471, 2022

    2022

  23. [23]

    The WU-Minn human connectome project: an overview,

    D. Van Essen, S. Smith, D. Barch, T. Behrens, E. Yacoub, K. Ugurbil, WU-Minn HCP Consortium,et al., “The WU-Minn human connectome project: an overview,”NeuroImage, vol. 80, pp. 62–79, 2013

    2013

  24. [24]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, W. Zaremba, V . Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,”Advances in neural information processing systems, vol. 29, 2016

    2016

  25. [25]

    Generative adversarial nets,

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” Advances in neural information processing systems, vol. 27, 2014

    2014

  26. [26]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,”Advances in neural information processing systems, vol. 30, 2017

    2017

  27. [27]

    Training generative adversarial networks with limited data,

    T. Karras, M. Aittala, J. Hellsten, S. Laine, J. Lehtinen, and T. Aila, “Training generative adversarial networks with limited data,”Advances in neural information processing systems, vol. 33, pp. 12104–12114, 2020. Fig. 2. Examples of generated pathological images with artifacts using LLIFT- GAN. From left to right column: Depicted are the original healt...

    2020

  28. [28]

    On leveraging pretrained GANs for generation with limited data,

    M. Zhao, Y . Cong, and L. Carin, “On leveraging pretrained GANs for generation with limited data,” inProceedings of the 37th International Conference on Machine Learning(H. D. III and A. Singh, eds.), vol. 119 ofProceedings of Machine Learning Research, pp. 11340– 11351, PMLR, 2020. APPENDIX Fig. 3. Examples of generated pathological images using LLIFT-DM...

    2020