Pith. sign in

REVIEW 4 major objections 5 minor 18 references

Spectral Regularization for Combating Mode Collapse in GANs

T0 review · 4 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 Compensating collapsed singular values in the discriminator's weights prevents GAN mode collapse, and spectral regularization does this in every setting where spectral-normalized GANs fail.

desk verdict Useful regularizer with an overblown causal story; the 'always outperform' claim fails on the paper's own FID numbers. read the letter →

arxiv 1908.10999 v3 pith:7MCCXQPC submitted 2019-08-29 cs.LG stat.ML

Kanglin Liu , Wenming Tang , Fei Zhou , Guoping Qiu This is my paper · ORCID
classification cs.LGstat.ML
keywords generativeadversarialnetworksmodecollapsespectralnormalizationregularizationsingularvaluesLipschitzconstraintimagegeneration
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 identifies a mechanism for mode collapse in spectral-normalized GANs and a way to switch it off. The authors monitor the singular values of the discriminator's weight matrices during training and find that collapse into a few output modes coincides with a sharp drop of many singular values, a pattern they call spectral collapse. They propose spectral regularization, which adds a compensation term that holds the smaller singular values up to the level of the largest one before spectral normalization is applied. In 26 settings on CIFAR-10, STL-10, and ImageNet, SR-GANs avoid mode collapse in every setting where SN-GANs collapse and report higher Inception Score and better Fréchet Inception Distance, so the method is a direct replacement candidate for spectral normalization.

What carries the argument

The central object is the spectral-regularization operator $W^{\mathrm{SR}}(W) = (W + \Delta W)/\sigma(W)$, a generalized normalization in which the weight matrix is first compensated along its singular directions and then divided by its spectral norm. It generalizes spectral normalization, which is the special case $\Delta W = 0$. The compensation is built from the singular-value decomposition $W = U\Sigma V^{T}$: static compensation raises the first $i$ singular values to the largest one, $\Delta\sigma_k = \sigma_1 - \sigma_k$, while dynamic compensation uses the historically maximal ratios $\gamma_j = \max(\sigma'_j/\sigma'_1)$ as targets. Corollary 1 supplies the reason this shape matters: a linear map satisfies the 1-Lipschitz constraint with equality only when all its singular values are 1, so a flat spectrum is the ideal and spectral collapse is a drift away from it. The gradient analysis then shows that the extra terms in $\partial W^{\mathrm{SR}}/\partial W_{ab}$ push the weight matrix to use all compensated singular directions rather than concentrating in one.

What would settle it

Find a single training run in which mode collapse occurs while the discriminator's singular values stay spread across many directions, or in which the spectrum is held flat by spectral regularization and mode collapse still occurs; either observation would refute the claim that preventing spectral collapse prevents mode collapse.

Watch

Extended reading notes

Core claim

Mode collapse in spectral-normalized GANs is accompanied by, and per the paper caused by, spectral collapse: while spectral normalization keeps the largest singular value of each discriminator layer at 1, the remaining singular values fall from near 1 to near 0 in the layers where collapse occurs. The paper's correction is spectral regularization, defined by $W^{\mathrm{SR}}(W) = (W+\Delta W)/\sigma(W)$, where $\Delta W = U\operatorname{diag}\{\Delta\sigma_1,\dots,\Delta\sigma_r,0,\dots,0\}V^{T}$ compensates the spectrum before the normalization step. Static compensation sets $\Delta\sigma_k = \sigma_1 - \sigma_k$ for the first $i$ singular values; dynamic compensation targets the largest ratio each singular value reached during training. The result, the paper reports, is that spectral regularization prevents mode collapse in all 10 settings where SN-GANs collapse and improves average Inception Score by 13.9% and FID by 21.8%.

Load-bearing premise

The load-bearing premise is that spectral collapse causes mode collapse rather than merely appearing alongside it, so compensating the singular values is sufficient to prevent mode collapse; the evidence is that the two phenomena always occur together and the compensation fixes both, which would also be true if spectral collapse were only a symptom of a deeper instability.

Editorial extensions

If this is right

  • SR-GANs avoid mode collapse in all 10 settings where SN-GANs collapse, including every small-channel group C setting, and the paper reports no mode collapse in any SR-GAN run.
  • Spectral regularization improves average Inception Score by 13.9% and FID by 21.8% over SN-GANs, with the largest gains on ImageNet conditional generation (44.9% IS improvement at setting E2048-64).
  • Because spectral normalization is the special case with no compensation, spectral regularization can replace spectral normalization inside existing architectures without changing the network or the optimization procedure.
  • Resuming a collapsing SN-GAN from its 40k-iteration snapshot with dynamic compensation prevents the collapse that would otherwise occur and improves IS by 44.9% and FID by 9.1%.
  • Static compensation suits low-resolution, few-category datasets, while dynamic compensation suits high-resolution, many-category datasets, and neither compensation scheme dominates across all comparisons.

Reading between the lines

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

  • Monitoring the full singular-value spectrum is a cheap early-warning test for mode collapse: because spectral collapse precedes or accompanies the quality drop in the paper's plots, one could trigger a correction as soon as the spectrum starts to fall, even before Inception Score declines.
  • The causal story is not fully settled by the paper's experiments, since no run induces spectral collapse independently while holding other optimization dynamics fixed; a direct intervention would separate cause from symptom.
  • The compensation recipe could be ported to other Lipschitz-constrained architectures, including the generator, or to other normalizations based on singular-value control, as a general anti-concentration penalty.
  • The group C results suggest small discriminator capacity is a strong collapse trigger; if spectral collapse is the bottleneck, spectral regularization may let practitioners use smaller discriminators or larger batches without the usual instability.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript studies mode collapse in GANs trained with spectral normalization (SN). It observes that when SN-GANs collapse, many singular values of the discriminator's weight matrices (beyond the top one, which is normalized to 1) drop near zero, a phenomenon the authors call spectral collapse. The paper argues that spectral collapse causes mode collapse and proposes spectral regularization (SR), which compensates these smaller singular values to keep them close to the largest one; two variants (static and dynamic compensation) are introduced. On a sweep of 26 settings over CIFAR-10, STL-10, and ImageNet, the paper reports that SR-GANs always avoid mode collapse in the 10 settings where SN-GANs collapse and that SR-GANs 'always outperform' SN-GANs in IS and FID. The authors also provide a theoretical justification via a Lipschitz-constraint corollary and a gradient analysis.

Significance. If the claims were correct, SR-GANs would be a simple and valuable drop-in replacement for spectral normalization. The paper's strengths include a relatively broad experimental sweep, public code, and the explicit identification of a recurring spectral signature (singular value drop) during collapse. However, the headline claim of always outperforming SN-GANs is internally falsified by one of the paper's own reported numbers, and the causal interpretation of spectral collapse is not established. The method might still be practically useful, but the manuscript currently overstates both the empirical and theoretical case.

major comments (4)
  1. [Section IV, Table II] The unqualified claim that 'in all cases, SR-GANs outperforms SN-GANs' and the abstract's 'not only always outperform' are contradicted by row A16-128: SN-GAN FID is 22.31±0.28 while SR-GAN FID is 24.67±0.28, i.e., SR is worse by 2.36 with non-overlapping error bars. This is a direct counterexample to the headline claim as written, unless the claim is explicitly qualified to Inception Score only or to settings with mode collapse. The paper should either remove 'always outperform' or add a clear metric- and setting-specific qualification and discuss the exception.
  2. [Section IV, first paragraph and Table II note] For the ten settings where SN-GANs collapse, Table II reports 'the best results before mode collapse' for SN-GANs, but no comparable selection rule is stated for SR-GANs (e.g., best-over-training vs. final iteration). If SN is evaluated at its best pre-collapse checkpoint and SR at its final checkpoint, the comparison is asymmetric and may bias the reported improvement. The authors should state the checkpoint/selection protocol for both methods and, ideally, also compare at matched training iterations.
  3. [Section III, first paragraph] The assertion that preventing spectral collapse 'demonstrat[es] that spectral collapse is the cause of mode collapse rather than a mere symptom' is not supported by the evidence presented. The paper shows that spectral collapse and mode collapse co-occur across 26 settings and that one intervention (SR), which changes multiple aspects of the optimization, prevents both. This is correlation plus a single intervention, not a causal demonstration; for example, no experiment induces spectral collapse independently while holding other dynamics fixed. The causal language should be softened to a hypothesis or association unless such an experiment is added.
  4. [Appendix A, Corollary 1] Corollary 1 is mathematically false as stated. For a linear map f(x)=Wx with ||Wx||≤||x||, the supremum of the Lipschitz constant is 1 and is achieved whenever the largest singular value of W equals 1, regardless of the remaining singular values (e.g., W=diag(1,0) achieves it). Requiring all singular values to be 1 is sufficient but not necessary. Since Corollary 1 is used to argue that the full spectral distribution matters, this theoretical justification needs to be corrected or replaced; the current statement overstates the theoretical role of the non-leading singular values.
minor comments (5)
  1. [Section II-C] The definition of spectral collapse is qualitative ('a large number of singular values drop dramatically'); a quantitative definition (e.g., threshold on the fraction of singular values below a fraction of σ_1, or the slope of the sorted singular value curve) would make the claimed association falsifiable and easier to reproduce.
  2. [Section IV, Table II] The number of independent runs used to compute the means and standard deviations in Table II is not stated; please report the number of seeds for IS and FID, and clarify whether error bars are standard deviations or standard errors.
  3. [Section II-C] There is a typo in Section II-C: 'ILRSVRC2012' should be 'ILSVRC2012'.
  4. [Section IV-C] In Section IV-C, the sentence 'our experiences is that setting i = 0.5r seems to work well' contains a grammatical error and should be rephrased.
  5. [Section III-A, Eq. (11)] The gradient formula in Eq. (11) is stated without derivation, and its notation (e.g., W_SN[u_1v_1^T]_{ab}) is not formally defined; since the training algorithm uses the explicit SVD compensation rather than this gradient, a derivation or a reference to where it is proved would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: spectral regularization is an externally evaluated intervention, not a fitted restatement of its inputs.

full rationale

The paper's central chain is: observe that spectral collapse and mode collapse co-occur in SN-GANs; hypothesize causality; design spectral regularization to keep singular values from collapsing; then test the method on external metrics (IS, FID, mode-collapse occurrence). The regularization target (compensating singular values toward sigma_1) is motivated by the observed healthy spectral distributions, but the headline outcomes are measured independently of the singular-value statistics. The only equation-level relation is W_SR(W) = W_SN(W) + DeltaW/sigma(W), which defines the method, and Corollary 1 is a genuine mathematical statement about Lipschitz constants; neither reduces mode-collapse prevention to the definition of spectral collapse. No load-bearing self-citation appears in the derivation chain: SN-GAN [10] and BigGAN [7] are used as external baselines, not as author self-citations. The causal claim that spectral collapse is 'the cause of mode collapse rather than a mere symptom' is under-supported—only one intervention, no independent manipulation of spectral collapse—but under-support is a correctness concern, not circularity. The internal contradiction in Table II, where setting A16-128 gives SN FID 22.31 +/- 0.28 versus SR FID 24.67 +/- 0.28, undermines the unqualified 'always outperform' headline, but that too is a falsification/correctness issue, not a circularity. The paper is self-contained against external benchmarks, so no circular step is exhibited.

Assumptions & free parameters 2 free parameters · 5 assumptions · 1 invented entities

The central claim rests on a small set of standard linear algebra facts, the empirical hypothesis that spectral collapse causes mode collapse, and the practical assumption that full or partial SVD of discriminator weights can be computed during training. The main free parameter (i=0.5r) and the per-dataset compensation choice are tuned to the reported benchmarks.

free parameters (2)
  • i (static compensation fraction) = 0.5r
    Section IV-C: 'The default value of the hyperparameter i for static compensation is empirically set as i = 0.5r.' The paper tests 0.25r, 0.5r, r and observes 0.5r works well, but the choice is tuned to the reported datasets.
  • compensation type per dataset = static for CIFAR-10/STL-10, dynamic for ImageNet
    Section IV-D: the choice of static vs dynamic compensation is made per dataset based on observed performance; the paper suggests static for low-resolution, few-category datasets and dynamic for high-resolution, many-category datasets. This is a hand-chosen, per-dataset selection.
assumptions (5)
  • standard math The Lipschitz constant of a linear map f(x)=Wx is its largest singular value sigma_1(W), and the bound ||Wx||<=||x|| is tight when sigma_1(W)=1.
    Used in Corollary 1 (Appendix A) to argue about the supremum of the Lipschitz constraint; the proof conflates the spectral norm bound with the condition that all singular values equal 1.
  • domain assumption Spectral distributions of D's weight matrices are computed via SVD W=U*Sigma*V^T and can be meaningfully compensated by adding DeltaW = U * DeltaD * V^T.
    Section III: the method relies on the ability to decompose each weight matrix and modify its singular values; no numerical or complexity analysis of computing full or partial SVD during training is provided.
  • ad hoc to paper Mode collapse in SN-GANs is caused by spectral collapse of the discriminator's weight matrices, not merely correlated with it.
    Section III states 'demonstrating that spectral collapse is the cause of mode collapse rather than a mere symptom.' This is the paper's central causal assumption and is not proven by controlled experiments.
  • domain assumption Inception Score and FID are reliable proxies for mode collapse and sample quality in the tested settings.
    Section II-C uses IS and FID to detect mode collapse; these are standard but the paper does not calibrate them against human evaluation for the small-channel and large-batch regimes.
  • ad hoc to paper The gradient formula in Eq. (11) correctly describes the derivative of W_SR with respect to W.
    Section III-A states the gradient 'we can write the gradient' without derivation; SVD-based gradients involve singular vector derivatives and are nontrivial.
invented entities (1)
  • spectral collapse independent evidence
    purpose: A diagnostic phenomenon: a large number of singular values of the spectral-normalized discriminator weight matrices drop dramatically when mode collapse occurs.
    The phenomenon is measurable from the SVD of weight matrices and is shown in plots (Figures 2-4), so it has an observable handle. But its causal role is inferred, not independently evidenced; the paper's own experiments only show correlation with mode collapse.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Spectral Regularization for Combating Mode Collapse in GANs." pith.science (2026). https://pith.science/paper/7MCCXQPC

@misc{pith2026190810999,
  author       = {Pith},
  title        = {Pith review of: Spectral Regularization for Combating Mode Collapse in GANs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7MCCXQPC}},
  note         = {Machine review of arXiv:1908.10999}
}
read the original abstract

Despite excellent progress in recent years, mode collapse remains a major unsolved problem in generative adversarial networks (GANs).In this paper, we present spectral regularization for GANs (SR-GANs), a new and robust method for combating the mode collapse problem in GANs. Theoretical analysis shows that the optimal solution to the discriminator has a strong relationship to the spectral distributions of the weight matrix.Therefore, we monitor the spectral distribution in the discriminator of spectral normalized GANs (SN-GANs), and discover a phenomenon which we refer to as spectral collapse, where a large number of singular values of the weight matrices drop dramatically when mode collapse occurs. We show that there are strong evidence linking mode collapse to spectral collapse; and based on this link, we set out to tackle spectral collapse as a surrogate of mode collapse. We have developed a spectral regularization method where we compensate the spectral distributions of the weight matrices to prevent them from collapsing, which in turn successfully prevents mode collapse in GANs. We provide theoretical explanations for why SR-GANs are more stable and can provide better performances than SN-GANs. We also present extensive experimental results and analysis to show that SR-GANs not only always outperform SN-GANs but also always succeed in combating mode collapse where SN-GANs fail. The code is available at https://github.com/max-liu-112/SRGANs-Spectral-Regularization-GANs-.

Figures

Figures reproduced from arXiv: 1908.10999 by the authors.

Figure 1
Figure 1. Conditional samples generated by our model base on ImageNet with setting [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Spectral distributions in the last layer for Good GANs (no mode collapse) at different number of iterations. The curves [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Spectral distributions in the last layer for settings where mode collapse occurs. The curves represent the spectral [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Spectral distributions (after 50k iterations) in layer 0 for different settings. (a) spectral distributions (b) Inception Score (c) Frechet Inception Distance ´ [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: An example showing how spectral distributions relate to Inception Score and Fr [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Inception Score, Frechet Inception Distance, spectral distribution and synthetic images of SN-GAN and SR-GAN for ´ the setting E2048−64. (a), (b) shows the IS and FID, respectively. (c) is the spectral distributions for layer 16 in SRGAN. (d) indicates the effect of SR…
Figure 7
Figure 7. Figure 7: Inception Score, Frechet Inception Distance, spectral distribution and synthetic images of SN-GAN and SR-GAN for ´ the setting C64−32 weight matrix. Readers are referred to Appendix for the details of the network architecture settings. When applying spectral regulariza…
Figure 8
Figure 8. Figure 8: The effect of i on model performance. N represents the number of singular values in corresponding weight matrix. (a) x ∼ qtrain (b) x ∼ qG (c) LD [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Statistics of D(x) and LD. (a) i=0.25r (b) i=0.50r (c) i=r [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 12
Figure 12. Figure 12: Comparison of two compensation methods. spectral distributions, and avoiding the occurrence of spectral collapse. As we can see in [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]
Figure 13
Figure 13. Figure 13: Architecture of D. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]
Figure 14
Figure 14. Figure 14: Spectral distributions settings B64−64. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 15
Figure 15. Figure 15: Spectral distributions settings B128−64. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p011_15.png]
Figure 16
Figure 16. Figure 16: Spectral distributions settings C8−32. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p011_16.png]
Figure 17
Figure 17. Figure 17: Spectral distributions settings C16−32 [PITH_FULL_IMAGE:figures/full_fig_p011_17.png]
Figure 18
Figure 18. Figure 18: Spectral distributions settings C32−32. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p012_18.png]
Figure 19
Figure 19. Figure 19: Spectral distributions settings C64−32. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p012_19.png]
Figure 20
Figure 20. Figure 20: Spectral distributions settings E256−64. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 [PITH_FULL_IMAGE:figures/full_fig_p012_20.png]
Figure 21
Figure 21. Figure 21: Spectral distributions settings E256−32. (a) layer 0 (b) layer 1 (c) layer 3 (d) layer 4 (e) layer 6 (f) layer 7 (g) layer 8 (h) layer 9 (i) layer 0 (j) layer 1 (k) layer 3 (l) layer 4 (m) layer 6 (n) layer 7 (o) layer 8 (p) layer 9 [PITH_FULL_IMAGE:figures/full_fig_…
Figure 22
Figure 22. Figure 22: Spectral distributions settings E2048−64 [PITH_FULL_IMAGE:figures/full_fig_p012_22.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 12 canonical work pages

  1. [1]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, and M. Mirza, “Generative adversarial nets,” Advances in neural information processing sys-tems , pp. 2672– 2680, 2014

    work page 2014

  2. [2]

    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, p. 1511.06434, 2015

    arXiv 2015

  3. [3]

    Wasserstein gan,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein gan,” arXiv preprint arXiv, p. 1701.07875, 2017

    arXiv 2017

  4. [4]

    Wasserstein Divergence for GANs

    J. Wu, Z. Huang, and J. Thoma, “Energy-relaxed wassertein gans (en- ergywgan): Towards more stable and high resolution image generation,” arXiv preprint arXiv , p. 1712.01026, 2017

    work page Pith review arXiv 2017

  5. [5]

    Began: Boundary equilibrium generative adversar-ial networks,

    D. Berthelot, T. Schumm, and L. Metz, “Began: Boundary equilibrium generative adversar-ial networks,” arXiv preprint arXiv , p. 1703.10717, 2017

    arXiv 2017

  6. [6]

    Least squares generative ad-versarial networks,

    X. Mao, Q. Li, and H. Xie, “Least squares generative ad-versarial networks,” 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2813–2821, 2017

    work page 2017

  7. [7]

    Large scale gan training for high fidelity natural image synthesis,

    B. Andrew, J. Donahue, and K. Simonyan, “Large scale gan training for high fidelity natural image synthesis,” arXiv preprint arXiv , 2018

    work page 2018

  8. [8]

    Improved training of wasser- stein gans,

    I. Gulrajani, F. Ahmed, and M. Arjovsky, “Improved training of wasser- stein gans,” Advances in Neural Information Processing Systems , pp. 5769–5779, 2017

    work page 2017

Show all 18 references
  1. [9]

    Loss-sensitive generative adversarial networks on lipschitz densities,

    G. Qi, “Loss-sensitive generative adversarial networks on lipschitz densities,” arXiv preprint arXiv , p. 1701.06264, 2017

    2017 arXiv

  2. [10]

    Spectral normalization for generative adversarial networks,

    T. Miyato, T. Kataoka, and M. Koyama, “Spectral normalization for generative adversarial networks,” arXiv preprint arXiv , p. 1802.05957, 2018

    2018 arXiv

  3. [11]

    Weight normalization: A simple reparam- eterization to accelerate training of deep neural networks,

    T. Salimans and D. Kingma, “Weight normalization: A simple reparam- eterization to accelerate training of deep neural networks,” Advances in Neural Information Processing Systems , pp. 901–909, 2016

    2016

  4. [12]

    Lectures on lipschitz analysis,

    J. Heinonen, “Lectures on lipschitz analysis,” University of Jyvaskyla , 2005

    2005

  5. [13]

    Neural photo editing with introspective adversarial networks,

    A. Brock, T. Lim, and J. M. Ritchie, “Neural photo editing with introspective adversarial networks,”arXiv preprint arXiv, p. 1609.07093, 2016

    2016 arXiv

  6. [14]

    80 million tiny images: A large data set for non-parametric object and scene recognition,

    A. Torralba, R. Fergus, and W. T. Freeman, “80 million tiny images: A large data set for non-parametric object and scene recognition,” IEEE Transactions on Pattern Analysis and Ma-chine Intelligence , vol. 30, no. 11, pp. 901–909, 2008

    2008

  7. [15]

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

    M. Heusel, H. Ramsauer, and T. Unterthiner, “Gans trained by a two time-scale update rule converge to a nash equilibrium,” arXiv preprint arXiv, p. 1706.08500, 2017

    2017 arXiv

  8. [16]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. S. L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” 2009 IEEE conference on computer vision and pattern recognition , pp. 248–255, 2009

    2009

  9. [17]

    cgans with projection discriminator,

    T. Miyato and M. Koyama, “cgans with projection discriminator,” arXiv preprint arXiv, p. 1808.05637, 2018

    2018 arXiv

  10. [18]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, and W. Zaremba, “Improved techniques for training gans,” Advances in Neural Information Processing Sys-tems , pp. 2234–2242, 2016. PUBLISHED AS A CONFERENCE PAPER ON ICCV 2019 11 (a) D architecture for CIFAR-10 and STL-10 (b) D architecture for Imag...

    2016

Pith tools

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