Pith. sign in

REVIEW 3 major objections 5 minor 59 references

BC-GAN: A Generative Adversarial Network for Synthesizing a Batch of Collocated Clothing

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read From one clothing image, BC-GAN generates a batch of visually collocated garments in the opposite category at once, using a frozen pre-trained StyleGAN to decode encoded style codes.

desk verdict The diversity headline is partly self-scored because the evaluation LPIPS is the same distance function the model is trained to maximize; the batch-generation idea is real, and the FID/compatibility evidence mostly holds up. read the letter →

arxiv 2502.01080 v1 pith:HJHEKXDC submitted 2025-02-03 cs.CV cs.MM

Dongliang Zhou , Haijun Zhang , Jianghong Ma , Jianyang Shi This is my paper · ORCID
classification cs.CVcs.MM
keywords generativeadversarialnetworkbatchclothinggenerationcollocatedsynthesisfashioncompatibilityimage-to-imagetranslationcontrastivelearningStyleGANinversionintelligence
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

BC-GAN is a generative adversarial network that takes a single clothing image and outputs several complementary garments—upper to lower or lower to upper—in one pass, where earlier collocated-clothing generators returned only one item and usually demanded extra attributes, preferences, or masks. The design sidesteps the spatial misalignment that defeats ordinary image-to-image translation by working in the latent space of a pre-trained StyleGAN: a random provisional target image is concatenated with the given garment, an encoder turns the pair into a style code, and the frozen StyleGAN decodes that code into a diverse but compatible item. Two new discriminators carry the supervision: a style-embedding discriminator that keeps encoded codes inside the StyleGAN $W$ distribution, and a contrastive compatibility discriminator that pulls synthesized pairs toward expert collocations and away from incompatible ones. On a new 31,631-outfit dataset, reported LPIPS, FID, and F2BT scores place BC-GAN ahead of five multimodal image-to-image baselines in both translation directions, and a user study reports majority preference for BC-GAN outputs over each baseline.

What carries the argument

The load-bearing object is the encoder–decoder path through StyleGAN's $W$ space, the internal latent space of a pre-trained StyleGAN generator. BC-GAN freezes the mapping network $f$ and synthesis network $g$ of a StyleGAN pre-trained on the target clothing domain, then trains an encoder $e$ (architecture based on pSp) that takes the channel-wise concatenation of the given garment image $x$ and a random provisional image $f\circ g(z_i)$ and outputs a style code $w_i$. Decoding $w_i$ via $g$ yields the final synthesized garment. This mechanism converts the spatially unaligned 'upper ⇌ lower' translation into a latent-space inference task: the provisional random image supplies the diversity, the concatenation with $x$ supplies the compatibility signal, and the $W$-space discriminator ensures the encoded code remains decodeable by the frozen generator. The contrastive compatibility discriminator and the LPIPS diversity loss then steer $w_i$ toward expert collocations and away from mode collapse.

What would settle it

Take held-out inputs from the DiverseOutfits test split and run the trained BC-GAN; then use the same F2BT predictor to compare each synthesized item against the actual expert-matched partner for that input, and in parallel run a forced-choice user study between generated outfits and true expert outfits. If true pairs do not consistently beat or match the generated pairs on both measures, the claimed fashion compatibility is an artifact of the evaluation rather than genuine collocation.

Watch

Extended reading notes

Core claim

The central claim is that batch collocated clothing synthesis can be reduced to an encoding problem in the latent space of a pre-trained StyleGAN, bypassing the pixel-level spatial alignment that ordinary image-to-image translation assumes. Given an input upper (or lower) garment $x$ and $n$ random codes, the generator computes a provisional image $f\circ g(z_i)$, concatenates it with $x$ in the channel dimension, and lets an encoder $e$ produce a style code $w_i$; decoding $w_i$ with the frozen synthesis network yields a garment that is both diverse across $i$ and compatible with $x$. The paper argues that existing multimodal image-to-image methods fail on 'upper ⇌ lower' because their encoder–decoder architectures presume aligned semantics, whereas the style-code route sidesteps alignment. A style-embedding discriminator keeps the encoded codes inside the StyleGAN $W$ distribution, and a contrastive compatibility discriminator, built as a Wasserstein-style adversarial critic, pulls synthesized pairs toward expert-annotated compatible pairs while pushing them away from incompatible and deliberately mismatched synthesized pairs. Experimental results reported for both translation directions show BC-GAN ahead of MUNIT, DRIT, DRIT++, StarGAN-v2, and SAVI2I on diversity (LPIPS), visual authenticity (FID), and fashion compatibility (F2BT), with an additional user study reporting preference rates of 76–96% over baselines.

Load-bearing premise

The method rests on the assumption that the pre-trained StyleGAN's internal code space is rich and well organized enough that the encoder's output code—not just a random code—decodes into a garment that genuinely matches the given one; Section IV-F notes that the random-to-encoded conversion is not easily disentangled for high-level compatibility learning in the lower-to-upper direction.

Editorial extensions

If this is right

  • A single input garment yields a set of candidate complementary garments in one forward pass, with no attribute annotations, masks, or user preferences as inputs.
  • The same generator serves both translation directions: pre-training the frozen StyleGAN on the target category and training the encoder for the opposite category gives 'upper → lower' or 'lower → upper' from the same code.
  • Style-space interpolation gives continuous control: mixing the random style code with the encoded style code at ratio $\alpha$ monotonically improves the measured fashion compatibility of the output.
  • Removing the contrastive compatibility discriminator drops F2BT from 22.9% to 14.9% in the upper-to-lower direction and from 22.8% to 17.0% in the lower-to-upper direction, so compatibility supervision is doing real work.
  • Removing the style-embedding discriminator, or replacing it with a pixel-level discriminator, raises FID from the reported 39–52 range to 260–339, indicating that latent-space supervision is what preserves visual authenticity.

Reading between the lines

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

  • Because the translation happens through latent codes rather than aligned pixels, the same recipe should transfer to other spatially unaligned generation tasks—for example, generating furniture that matches a rug from a room photo—whenever a target-domain StyleGAN can be pre-trained.
  • The paper's Section IV-F admission that the random-to-encoded style conversion is not easily disentangled for high-level compatibility learning in the lower-to-upper direction suggests the method's ceiling is set by $W$-space disentanglement; a direct probe would be to repeat the experiments with a more strongly regularized StyleGAN variant and check whether F2BT improves.
  • The contrastive compatibility discriminator samples incompatible pairs by random combination; replacing them with hard negatives mined by a learned compatibility ranker is a natural extension the paper does not explore and could raise the ceiling further.
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

3 major / 5 minor

Summary. The paper proposes BC-GAN, a generative adversarial network for synthesizing a batch of visually collocated clothing items from a single input clothing image, for both the upper-to-lower and lower-to-upper translation directions. The generator uses a frozen pre-trained StyleGAN: random latent codes are mapped to target-domain images, a learned encoder maps the concatenation of the input image and a random synthetic target image into a W-space style code, and the frozen synthesis network decodes that code into the output. Training combines a style-embedding adversarial loss, a contrastive-learning compatibility discriminator, and an explicit LPIPS-based diversity loss. The authors construct a dataset called DiverseOutfits from Polyvore and compare BC-GAN against MUNIT, DRIT, DRIT++, StarGAN-v2, and SAVI2I using LPIPS diversity, FID, F2BT compatibility, and a user study.

Significance. If the claims are validated, BC-GAN would be the first framework to synthesize multiple collocated clothing images simultaneously and would offer a useful way to adapt frozen StyleGAN latent spaces to cross-category fashion translation. The paper has several strengths: the problem formulation is clear, the architecture is simple and avoids spatial alignment assumptions by operating in a pre-trained latent space, the contrastive compatibility discriminator is a sensible way to exploit real/incompatible/fake pairs, and the paper includes ablations for each loss component. However, the headline diversity claim is currently confounded because the training objective and the evaluation metric are the same LPIPS distance, and the compatibility evidence rests on a cross-dataset predictor and a small user study without uncertainty quantification. These issues are fixable with additional experiments, but they are load-bearing for the paper's central conclusions.

major comments (3)
  1. [§III-D, Eq. (5); §IV-C1, Eq. (7); Tables I, II, IV] The diversity evaluation is not independent of the training objective. The generator's diversity loss in Eq. (5) is Ldiv = −E[d(yi, yj)] with d implemented by LPIPS, and the evaluation metric in Eq. (7) averages exactly the same LPIPS distance d over synthesized pairs. Because BC-GAN is explicitly trained to maximize this quantity while the compared baselines are not, the LPIPS margins in Tables I and II are expected even if the outputs are not more perceptually varied. The ablation in Table IV demonstrates only that optimizing Ldiv moves the LPIPS score; it does not establish that the diversity gain is visually meaningful or that it transfers to an independent diversity measure. Please add at least one diversity metric that is not optimized during training, such as LPIPS with a different backbone network, attribute statistics, or a human ranking of distinctiveness, and report whether the relative ordering of methods persists.
  2. [§IV-C3, Eq. (8); Tables I, II] The F2BT compatibility scores are reported without any measure of uncertainty. The predictor φ used in Eq. (8) was trained on the Maryland Polyvore dataset, which differs from the DiverseOutfits test set, and the per-outfit comparisons are aggregated into a single point estimate. Since the headline advantage over the second-best method is 2.9 percentage points in Table I and 6.2 points in Table II, it is important to report bootstrap confidence intervals, paired tests, or repeated-seed variances, and to quantify the predictor's calibration on DiverseOutfits outfits. Without this information, the compatibility claim rests on a single cross-dataset predictor with unknown precision; the same concern applies to the FID and LPIPS point estimates in Tables I and II.
  3. [§IV-D, Table III] The user study involves only 10 participants and reports raw preference percentages without inter-annotator agreement, per-participant variance, or significance testing. The 'BC-GAN > Real' row (40.7%) shows that expert-constructed outfits are preferred over BC-GAN outputs more than half the time, so the statement in §IV-D that the compatibility of synthesized items 'closely parallels' expert-constructed outfits requires a statistical comparison rather than a single point estimate. Please report the distribution across participants and test the results against the null hypothesis of equal preference; this is necessary to support the compatibility claim beyond the F2BT metric.
minor comments (5)
  1. [Algorithm 1, lines 16–19] The pseudocode is inconsistent with the feed-forward description in Section III-C. Line 17 writes yi ← e ◦ g(x ⊕ g(wi)), but the correct forward pass should be yi ← g(e(x ⊕ g(worig_i))); moreover, line 19 computes Ladv using Dse(yi), although Dse operates on style embeddings and should receive wi or e(...). Please correct these notation errors, as they make the exact training procedure ambiguous.
  2. [§IV-E, discussion of Table VI and Table VII] The text says 'In Table VI, BC-GAN w/o contrastive learning indicates...' but the corresponding result appears in Table VII; Table VI contains the comparison without Dcmp. Please fix the cross-reference.
  3. [§IV-A, dataset construction] The dataset construction uses an LPIPS cut-off to merge duplicate items, but no sensitivity analysis for this threshold is provided, and no dataset release link is given. Please report the chosen cut-off and its effect on the number of distinct items, and consider releasing the dataset to support reproducibility.
  4. [§III-D, Eq. (6)] The loss weights λ1 = 1 and λ2 = 3 are described as empirically set, but no sensitivity study is reported. Since Ldiv, Ladv, and Lcmp operate on very different scales, a brief ablation or a plot over λ values would strengthen the claim that the chosen balance is not responsible for the reported improvements.
  5. [§IV-B, baselines] The baselines are said to use the authors' original implementations and are trained from scratch, but no hyperparameter settings or training durations are reported for them. A brief table of training configurations would improve the fairness of the comparison.

Circularity Check

1 steps flagged · score 6.0 of 10

Diversity claim is confounded: the LPIPS evaluation metric (Eq. 7) is identical to the LPIPS diversity loss the generator maximizes (Eq. 5), so BC-GAN's reported diversity advantage is partially self-scored by construction.

  1. fitted input called prediction [Section III-D Eq. (5) and Section IV-C Eq. (7)]
    "Ldiv = −Eyi̸=yj [d(yi, yj)], (5) ... In our implementation, we used the LPIPS [42] as our distance metric. ... LPIPS =Ex∼X [ NX i=1 NX j=i+1 2 × d(yi, yj) N × (N − 1) ], (7) ... The distance function d(·, ·) was implemented by LPIPS with a pre-trained AlexNet to evaluate the performance in terms of diversity."

    The evaluation metric in Eq. (7) is exactly the same pairwise LPIPS distance d(·,·) that Eq. (5) maximizes as the training objective Ldiv = −E_{yi≠yj}[d(yi,yj)]. BC-GAN's generator is explicitly optimized to increase LPIPS distance among its outputs, while the baselines (MUNIT, DRIT, DRIT++, StarGAN-v2, SAVI2I) are not trained with this objective. The LPIPS improvements reported in Tables I and II, and the ablation gain in Table IV, are therefore largely a check that the loss moves its own metric rather than an independent measurement of perceptual diversity. The diversity component of the headline claim is self-scored by construction.

full rationale

BC-GAN's central architecture is otherwise self-contained: the adversarial objective supervises encoded style embeddings against the frozen StyleGAN latent distribution, the compatibility discriminator is trained against real/compatible and real/incompatible pairs, and FID uses an external Inception feature extractor. The F2BT metric is also externally grounded because the predictor φ comes from MMFashion and was trained on the Maryland Polyvore dataset, not on the authors' DiverseOutfits, so the self-citation to Zhou et al. [18] for F2BT is not load-bearing. The paper's own Section IV-F caveat that the conversion from worig_i to wi 'cannot be easily disentangled for a high-level learning, i.e., fashion compatibility learning' is an honest robustness limitation rather than a circular step. The one clear circular step is the diversity evaluation: Section III-D Eq. (5) defines the diversity loss as negative pairwise LPIPS distance, and Section IV-C Eq. (7) defines the diversity metric as the same pairwise LPIPS distance. Because BC-GAN directly optimizes this metric while the baselines do not, its LPIPS advantage is substantially self-scored. This is partial circularity, not total, because the authenticity and compatibility claims rest on independent measurements. Additional reporting concerns such as the absence of error bars and the self-constructed dataset are correctness risks, not circularity.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on assumptions about ground-truth compatibility labels from expert-curated Polyvore outfits and about the expressiveness and disentanglement of the frozen StyleGAN latent space. The loss weights, dataset construction threshold, and the choice of LPIPS as the diversity distance are hand-set and directly affect the reported results.

free parameters (5)
  • loss weights lambda1 and lambda2 = 1 and 3
    Set empirically in Section IV-B to balance diversity and compatibility losses in Eq. (6).
  • LPIPS cut-off for merging duplicate items in dataset = not specified numerically
    Used in Section IV-A to merge visually identical clothing entities; the threshold is hand-chosen and affects dataset statistics.
  • training hyperparameters = batch size 4, 50,000 iterations, learning rate 2e-4, Adam beta1=0, beta2=0.99
    Listed in Section IV-B; standard but hand-set values that influence final model quality.
  • discriminator pool size = 100
    Used for the style embedding discriminator pool in Section IV-B to stabilize training.
  • diversity distance function = LPIPS (AlexNet)
    Chosen because L1/L2 alternatives failed to converge, as stated in Section III-D; this choice directly ties the training objective to the evaluation metric.
assumptions (4)
  • domain assumption Expert-curated Polyvore outfits are valid ground-truth compatible pairs, and random pairs are incompatible.
    Used to construct the training labels for the compatibility discriminator in Sections III-C and IV-A.
  • domain assumption The frozen pre-trained StyleGAN's W space is expressive and disentangled enough for the encoder to encode compatibility and for the synthesis network to decode it.
    Core mechanism of the generator in Section III-B; the authors themselves note imperfect disentanglement in Section IV-F.
  • domain assumption The F2BT predictor trained on Maryland Polyvore transfers to evaluating outfits from DiverseOutfits.
    Used as the external fashion compatibility metric in Section IV-C, despite the domain difference.
  • standard math WGAN training and contrastive learning converge to useful solutions.
    Standard assumptions for adversarial training; no convergence proof is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BC-GAN: A Generative Adversarial Network for Synthesizing a Batch of Collocated Clothing." pith.science (2026). https://pith.science/paper/HJHEKXDC

@misc{pith2026250201080,
  author       = {Pith},
  title        = {Pith review of: BC-GAN: A Generative Adversarial Network for Synthesizing a Batch of Collocated Clothing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HJHEKXDC}},
  note         = {Machine review of arXiv:2502.01080}
}
read the original abstract

Collocated clothing synthesis using generative networks has become an emerging topic in the field of fashion intelligence, as it has significant potential economic value to increase revenue in the fashion industry. In previous studies, several works have attempted to synthesize visually-collocated clothing based on a given clothing item using generative adversarial networks (GANs) with promising results. These works, however, can only accomplish the synthesis of one collocated clothing item each time. Nevertheless, users may require different clothing items to meet their multiple choices due to their personal tastes and different dressing scenarios. To address this limitation, we introduce a novel batch clothing generation framework, named BC-GAN, which is able to synthesize multiple visually-collocated clothing images simultaneously. In particular, to further improve the fashion compatibility of synthetic results, BC-GAN proposes a new fashion compatibility discriminator in a contrastive learning perspective by fully exploiting the collocation relationship among all clothing items. Our model was examined in a large-scale dataset with compatible outfits constructed by ourselves. Extensive experiment results confirmed the effectiveness of our proposed BC-GAN in comparison to state-of-the-art methods in terms of diversity, visual authenticity, and fashion compatibility.

Figures

Figures reproduced from arXiv: 2502.01080 by the authors.

Figure 1
Figure 1. Multiple visually-collocated fashion items synthesized by our BC [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparable frameworks of the StyleGAN-based models: (a) StyleGAN [36], (b) GAN inversion for edit framework [29], and (c) BC-GAN (ours). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Key components of our proposed BC-GAN: (a) The generator [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Statistics of our DiverseOutfits dataset: (a) an illustration of multiple [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Comparisons between our BC-GAN and other multimodal I2I translation methods, which are MUNIT [19], DRIT [20], DRIT++ [21], StarGAN-v2 [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Visual comparison results for the effectiveness of diversity loss: (a) [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Visual comparison results for the effectiveness of compatibility [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Linear interpolation between w orig i and wi, where w orig i is a style embedding mapped from a random latent code into W space, and wi is a style embedding encoded from a given clothing image and a random synthesized image into W space: (a) continuous translation on t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 56 canonical work pages

  1. [1]

    Image-based recommendations on styles and substi- tutes,

    J. McAuley et al., “Image-based recommendations on styles and substi- tutes,” in Proc. of SIGIR , 2015, pp. 43–52

    work page 2015

  2. [2]

    Learning visual clothing style with heterogeneous dyadic co-occurrences,

    A. Veit et al., “Learning visual clothing style with heterogeneous dyadic co-occurrences,” in Proc. of ICCV , 2015, pp. 4642–4650. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 14

    work page 2015

  3. [3]

    Learning fashion compatibility with bidirectional LSTMs,

    X. Han et al. , “Learning fashion compatibility with bidirectional LSTMs,” in Proc. of ACMMM, 2017, pp. 1078–1086

    work page 2017

  4. [4]

    Learning type-aware embeddings for fashion compatibility,

    M. I. Vasileva et al. , “Learning type-aware embeddings for fashion compatibility,” in Proc. of ECCV , 2018, pp. 390–405

    work page 2018

  5. [5]

    Dressing as a whole: Outfit compatibility learning based on node-wise graph neural networks,

    Z. Cui et al., “Dressing as a whole: Outfit compatibility learning based on node-wise graph neural networks,” in Proc. of WWW, 2019, pp. 307– 317

    work page 2019

  6. [6]

    Hierarchical fashion graph network for personalized outfit recommendation,

    X. Li et al., “Hierarchical fashion graph network for personalized outfit recommendation,” in Proc. of SIGIR , 2020, pp. 159–168

    work page 2020

  7. [7]

    Generative adversarial nets,

    I. Goodfellow et al., “Generative adversarial nets,” in Proc. of NeurIPS, vol. 27, 2014, pp. 1–9

    work page 2014

  8. [8]

    Spatiotemporal generative adversarial network-based dynamic texture synthesis for surveillance video coding,

    K. Yang et al. , “Spatiotemporal generative adversarial network-based dynamic texture synthesis for surveillance video coding,” IEEE Trans. on Circuits and Syst. for Video Technol. , vol. 32, no. 1, pp. 359–373, 2021

    work page 2021

Show all 59 references
  1. [9]

    Texture brush for fashion inspiration transfer: A gen- erative adversarial network with heatmap-guided semantic disentangle- ment,

    H. Yan et al. , “Texture brush for fashion inspiration transfer: A gen- erative adversarial network with heatmap-guided semantic disentangle- ment,” IEEE Trans. on Circuits and Syst. for Video Technol. , 2022

    2022

  2. [10]

    Local and global perception generative adversarial network for facial expression synthesis,

    Y . Xia et al. , “Local and global perception generative adversarial network for facial expression synthesis,” IEEE Trans. on Circuits and Syst. for Video Technol., vol. 32, no. 3, pp. 1443–1452, 2021

    2021

  3. [11]

    Learning deep patch representation for probabilistic graphical model-based face sketch synthesis,

    M. Zhu et al. , “Learning deep patch representation for probabilistic graphical model-based face sketch synthesis,” International Journal of Computer Vision, vol. 129, pp. 1820–1836, 2021

    2021

  4. [12]

    Knowledge distillation for face photo–sketch synthesis,

    ——, “Knowledge distillation for face photo–sketch synthesis,” IEEE Transactions on Neural Networks and Learning Systems , vol. 33, no. 2, pp. 893–906, 2020

    2020

  5. [13]

    A deep collaborative framework for face photo–sketch synthesis,

    ——, “A deep collaborative framework for face photo–sketch synthesis,” IEEE transactions on neural networks and learning systems , vol. 30, no. 10, pp. 3096–3108, 2019

    2019

  6. [14]

    Toward AI fashion design: An Attribute-GAN model for clothing match,

    L. Liu et al., “Toward AI fashion design: An Attribute-GAN model for clothing match,” Neurocomputing, vol. 341, 2019

    2019

  7. [15]

    Collocating clothes with generative adversarial networks cosu- pervised by categories and attributes: A multidiscriminator framework,

    ——, “Collocating clothes with generative adversarial networks cosu- pervised by categories and attributes: A multidiscriminator framework,” IEEE Trans. on Neural Netw. and Learning Syst. , vol. 31, no. 9, pp. 3540–3554, 2019

    2019

  8. [16]

    Personalized fashion design,

    C. Yu et al., “Personalized fashion design,” in Proc. of ICCV, 2019, pp. 9046–9055

    2019

  9. [17]

    Learning to synthesize compatible fashion items using semantic alignment and collocation classification: An outfit generation framework,

    D. Zhou et al., “Learning to synthesize compatible fashion items using semantic alignment and collocation classification: An outfit generation framework,” IEEE Trans. on Neural Netw. and Learning Syst., pp. 1–16, 2023

    2023

  10. [18]

    COutfitGAN: Learning to synthesize compatible outfits super- vised by silhouette masks and fashion styles,

    ——, “COutfitGAN: Learning to synthesize compatible outfits super- vised by silhouette masks and fashion styles,” IEEE Trans. on Multime- dia, pp. 1–15, 2023

    2023

  11. [19]

    Multimodal unsupervised image-to-image translation,

    X. Huang et al., “Multimodal unsupervised image-to-image translation,” in Proc. of ECCV , 2018, pp. 172–189

    2018

  12. [20]

    Diverse image-to-image translation via disentangled representations,

    H.-Y . Lee et al. , “Diverse image-to-image translation via disentangled representations,” in Proc. of ECCV , 2018, pp. 35–51

    2018

  13. [21]

    DRIT++: Diverse image-to-image translation via disentangled representations,

    ——, “DRIT++: Diverse image-to-image translation via disentangled representations,” Int. Journal of Comput. Vis., vol. 128, no. 10, pp. 2402– 2417, 2020

    2020

  14. [22]

    StarGAN v2: Diverse image synthesis for multiple domains,

    Y . Choi et al. , “StarGAN v2: Diverse image synthesis for multiple domains,” in Proc. of CVPR , 2020, pp. 8185–8194

    2020

  15. [23]

    Continuous and diverse image-to-image translation via signed attribute vectors,

    Q. Mao et al., “Continuous and diverse image-to-image translation via signed attribute vectors,” Int. Journal of Comput. Vis. , vol. 130, no. 2, pp. 517–549, 2022

    2022

  16. [24]

    SCGAN: Saliency map-guided colorization with gener- ative adversarial network,

    Y . Zhao et al., “SCGAN: Saliency map-guided colorization with gener- ative adversarial network,” IEEE Trans. on Circuits and Syst. for Video Technol., vol. 31, no. 8, pp. 3062–3077, 2020

    2020

  17. [25]

    Dual-domain generative adversarial network for digital image operation anti-forensics,

    H. Xie, J. Ni, and Y .-Q. Shi, “Dual-domain generative adversarial network for digital image operation anti-forensics,” IEEE Trans. on Circuits and Syst. for Video Technol. , vol. 32, no. 3, pp. 1701–1706, 2021

    2021

  18. [26]

    A robust coverless steganography based on generative adversarial networks and gradient descent approximation,

    F. Peng, G. Chen, and M. Long, “A robust coverless steganography based on generative adversarial networks and gradient descent approximation,” IEEE Trans. on Circuits and Syst. for Video Technol. , vol. 32, no. 9, pp. 5817–5829, 2022

    2022

  19. [27]

    Learning a similarity metric discriminatively, with application to face verification,

    S. Chopra, R. Hadsell, and Y . LeCun, “Learning a similarity metric discriminatively, with application to face verification,” inProc. of CVPR, vol. 1. IEEE, 2005, pp. 539–546

    2005

  20. [28]

    Bidirectional recurrent neural networks,

    M. Schuster and K. Paliwal, “Bidirectional recurrent neural networks,” IEEE Trans. on Signal Process. , vol. 45, 1997

    1997

  21. [29]

    In-domain GAN inversion for real image editing,

    J. Zhu et al., “In-domain GAN inversion for real image editing,” in Proc. of ECCV. Springer, 2020, pp. 592–608

    2020

  22. [30]

    Encoding in style: A StyleGAN encoder for image-to-image translation,

    E. Richardson et al. , “Encoding in style: A StyleGAN encoder for image-to-image translation,” in Proc. of CVPR , 2021, pp. 2287–2296

    2021

  23. [31]

    Designing an encoder for StyleGAN image manipulation,

    O. Tov et al., “Designing an encoder for StyleGAN image manipulation,” ACM Trans. on Graphics , vol. 40, no. 4, pp. 1–14, 2021

    2021

  24. [32]

    Gan inversion: A survey,

    W. Xia et al., “Gan inversion: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 3, pp. 3121–3138, 2022

    2022

  25. [33]

    Stylerig: Rigging stylegan for 3d control over portrait images,

    A. Tewari et al., “Stylerig: Rigging stylegan for 3d control over portrait images,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2020, pp. 6142–6151

    2020

  26. [34]

    Seeing what a gan cannot generate,

    D. Bau et al., “Seeing what a gan cannot generate,” in Proceedings of the IEEE/CVF International Conference on Computer Vision , 2019, pp. 4502–4511

    2019

  27. [35]

    Interfacegan: Interpreting the disentangled face repre- sentation learned by gans,

    Y . Shen et al. , “Interfacegan: Interpreting the disentangled face repre- sentation learned by gans,” TPAMI, 2020

    2020

  28. [36]

    A style-based generator architecture for generative adversarial networks,

    T. Karras, S. Laine, and T. Aila, “A style-based generator architecture for generative adversarial networks,” in Proc. of CVPR, 2019, pp. 4401– 4410

    2019

  29. [37]

    Fcboost-net: A generative network for synthesizing multiple collocated outfits via fashion compatibility boosting,

    D. Zhou et al. , “Fcboost-net: A generative network for synthesizing multiple collocated outfits via fashion compatibility boosting,” in Proc. of ACMMM. ACM, 2023, pp. 1–9

    2023

  30. [38]

    Analyzing and improving the image quality of StyleGAN,

    T. Karras et al. , “Analyzing and improving the image quality of StyleGAN,” in Proc. of CVPR , 2020, pp. 8110–8119

    2020

  31. [39]

    Training generative adversarial networks with limited data,

    ——, “Training generative adversarial networks with limited data,” in Proc. of NeurIPS , 2020, pp. 12 104–12 114

    2020

  32. [40]

    Auto-encoding variational Bayes,

    D. P. Kingma and M. Welling, “Auto-encoding variational Bayes,” in Proc. of ICLR , 2014, pp. 1–14

    2014

  33. [41]

    Image-to-image translation with conditional adversarial networks,

    P. Isola et al., “Image-to-image translation with conditional adversarial networks,” in Proc. of CVPR , 2017, pp. 1125–1134

    2017

  34. [42]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang et al. , “The unreasonable effectiveness of deep features as a perceptual metric,” in Proc. of CVPR , 2018

    2018

  35. [43]

    Arcface: Additive angular margin loss for deep face recognition,

    J. Deng et al. , “Arcface: Additive angular margin loss for deep face recognition,” in Proc. of CVPR , 2019, pp. 4690–4699

    2019

  36. [44]

    Dimensionality reduction by learning an invariant mapping,

    R. Hadsell, S. Chopra, and Y . LeCun, “Dimensionality reduction by learning an invariant mapping,” in Proc. of CVPR, vol. 2. IEEE, 2006, pp. 1735–1742

    2006

  37. [45]

    Facenet: A unified embed- ding for face recognition and clustering,

    F. Schroff, D. Kalenichenko, and J. Philbin, “Facenet: A unified embed- ding for face recognition and clustering,” in Proc. of CVPR , 2015, pp. 815–823

    2015

  38. [46]

    A simple framework for contrastive learning of visual representations,

    T. Chen et al. , “A simple framework for contrastive learning of visual representations,” in Proc. of ICML . PMLR, 2020, pp. 1597–1607

    2020

  39. [47]

    Spatiotemporal decouple-and-squeeze contrastive learning for semisupervised skeleton-based action recognition,

    B. Xu et al., “Spatiotemporal decouple-and-squeeze contrastive learning for semisupervised skeleton-based action recognition,” IEEE Transac- tions on Neural Networks and Learning Systems , 2023

    2023

  40. [48]

    Pyramid self-attention polymerization learning for semi-supervised skeleton-based action recognition,

    B. Xu and X. Shu, “Pyramid self-attention polymerization learning for semi-supervised skeleton-based action recognition,” arXiv preprint arXiv:2302.02327, 2023

    2023 arXiv

  41. [49]

    Progressive instance-aware feature learning for compo- sitional action recognition,

    R. Yan et al. , “Progressive instance-aware feature learning for compo- sitional action recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

    2023

  42. [50]

    Multi-granularity anchor-contrastive representation learn- ing for semi-supervised skeleton-based action recognition,

    X. Shu et al., “Multi-granularity anchor-contrastive representation learn- ing for semi-supervised skeleton-based action recognition,” IEEE Trans- actions on Pattern Analysis and Machine Intelligence , 2022

    2022

  43. [51]

    X-invariant contrastive augmentation and representation learning for semi-supervised skeleton-based action recognition,

    B. Xu, X. Shu, and Y . Song, “X-invariant contrastive augmentation and representation learning for semi-supervised skeleton-based action recognition,” IEEE Transactions on Image Proces. , vol. 31, pp. 3852– 3867, 2022

    2022

  44. [52]

    Wasserstein generative adver- sarial networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative adver- sarial networks,” in Proc. of ICML . PMLR, 2017, pp. 214–223

    2017

  45. [53]

    Fine-grained visual comparisons with local learning,

    A. Yu and K. Grauman, “Fine-grained visual comparisons with local learning,” in Proc. of CVPR , 2014, pp. 192–199

    2014

  46. [54]

    Automatic differentiation in PyTorch,

    A. Paszke et al. , “Automatic differentiation in PyTorch,” in Proc. of NeurIPS, 2017, pp. 1–4

    2017

  47. [55]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. of ICLR , 2014, pp. 1–15

    2014

  48. [56]

    GANs trained by a two time-scale update rule converge to a local Nash equilibrium,

    M. Heusel et al. , “GANs trained by a two time-scale update rule converge to a local Nash equilibrium,” in Proc. of NeurIPS , 2017, pp. 1–12

    2017

  49. [57]

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

    Y . Choi et al. , “StarGAN: Unified generative adversarial networks for multi-domain image-to-image translation,” in Proc. of CVPR, 2018, pp. 8789–8797

    2018

  50. [58]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Proc. of NeurIPS , vol. 33, pp. 6840–6851, 2020

    2020

  51. [59]

    High-resolution image synthesis with latent diffu- sion models,

    R. Rombach et al., “High-resolution image synthesis with latent diffu- sion models,” in Proc. of CVPR , 2022, pp. 10 684–10 695

    2022

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.