Pith. sign in

REVIEW 4 major objections 6 minor 27 references

End-to-end Training for Text-to-Image Synthesis using Dual-Text Embeddings

T0 review · 4 major / 6 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 Splitting text embeddings into a realism branch and an alignment branch improves text-to-image GANs.

desk verdict Decoupling generator- and discriminator-side text embeddings is a plausible idea with a solid internal ablation, but the circular R-precision metric means the alignment claim doesn't hold up; the photo-realism gains do. read the letter →

arxiv 2502.01507 v1 pith:KMX6ULIB submitted 2025-02-03 cs.CV

Yeruru Asrar Ahmed , Anurag Mittal This is my paper · ORCID
classification cs.CV
keywords text-to-imagesynthesisdualtextembeddingsend-to-endtraininggenerativeadversarialnetworkscontrastivelearningphoto-realismtext-imagealignment
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 proposes that text embeddings for text-to-image synthesis should be learned inside the synthesis network, rather than borrowed from a pre-trained, generically trained text encoder. Its concrete claim is that two separate text embeddings work better than one shared embedding: a generator-side embedding trained to make generated images photo-realistic, and a discriminator-side embedding trained to keep images aligned with their captions. The paper implements this as DTE-GAN, a single-stage GAN with two Bi-LSTM text encoders, and reports lower Fréchet Inception Distance on CUB and Oxford-102 than prior GAN approaches that use pre-trained embeddings, and comparable MS-COCO results with far fewer parameters. The authors also show the learned embeddings can be reused for text-to-image manipulation. If the attribution holds, the useful principle is to decouple 'make it look real' from 'match the text' as separate conditioning representations.

What carries the argument

The central object is the dual text embedding setup: two independent word-embedding layers ($W_G$ and $W_D$) with two Bi-LSTM sentence encoders, one producing the generator-side sentence embedding $S_G$ and the other the discriminator-side sentence embedding $S_D$. The mechanism that makes the split work is gradient isolation: $S_G$ learns only from generation losses (adversarial loss plus multi-modal contrastive loss on fake images), while $S_D$ learns only from the multi-modal contrastive loss on real image-text pairs, with a stop-gradient preventing noisy generation gradients from flowing into $S_D$ when it is passed to the generator.

What would settle it

Run a strictly controlled comparison where the architecture is held fixed and only the embedding scheme changes—same generator and discriminator widths, losses, training schedule, and random seeds, with dual embeddings versus a single shared embedding—and test whether the dual version still improves FID and R-precision; alternatively, insert DTE-GAN's dual embeddings into DF-GAN's exact generator and discriminator and check whether FID improves over DF-GAN's reported 14.81 on CUB. If the gap disappears in either test, the dual-embedding claim is not responsible for the published advantage.

Watch

Extended reading notes

Core claim

The central claim is that decoupling text conditioning into two independently trained embeddings improves text-to-image generation. DTE-GAN uses separate word embeddings and Bi-LSTM sentence encoders for the generator and the discriminator: the generator-side embedding is updated only by the adversarial loss and by a contrastive loss computed on generated images, while the discriminator-side embedding is updated only by a contrastive loss computed on real image-text pairs. The generator additionally receives a detached copy of the discriminator-side sentence embedding, which the authors call a 'sneak peek' at a stable, alignment-focused representation. The paper reports that a shared embedding trained with both losses produces worse images, and that feeding the noisy generator-side embedding to the discriminator degrades performance, which it reads as evidence that separate, purpose-specific embeddings are the effective design choice.

Load-bearing premise

The conclusion that the dual-embedding design itself causes the reported gains rests on ablations performed inside DTE-GAN's own architecture; comparisons against earlier methods change the generator, discriminator, and parameter count simultaneously, so the dual embeddings may not be the actual source of the improvement over those baselines.

Editorial extensions

If this is right

  • On CUB, DTE-GAN lowers FID to 13.67 and raises R-precision to 86.64, beating DF-GAN (14.81 FID) and AttnGAN (23.98 FID) on the same dataset.
  • On Oxford-102, DTE-GAN lowers FID to 30.07 from DualAttn-GAN's 40.31 and raises Inception Score to 4.21.
  • On MS-COCO, DTE-GAN reaches FID 25.17 with 11M parameters, and 19.69 with DTE-GAN+MAGP, matching DF-GAN and SSA-GAN while using roughly half their parameter count.
  • Table 6 shows the dual-embedding configuration beats all shared-embedding variants of the same architecture on IS, FID, and R-precision.
  • The learned dual embeddings transfer: used as fixed word embeddings in a lightweight text-guided manipulation GAN, they improve FID from 8.02 to 7.77 on CUB.

Reading between the lines

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

  • If the decoupling principle generalizes, frozen pre-trained text encoders like CLIP may be leaving generative performance on the table: they are optimized for alignment and retrieval, not for driving a decoder toward realism. A testable extension would be to add a trainable realism branch to a diffusion-based text-to-image model while keeping a frozen alignment branch.
  • The successful 'sneak peek'—feeding the generator a detached, contrastively trained sentence embedding—suggests a general design rule for conditional generative models: the cleanest representation of the condition is not necessarily the one that should be fine-tuned by the generator's own noisy gradients. This could be probed in other conditional GANs by adding or removing the stop-gradient on the
  • The same dual-embedding split could transfer to other multimodal generation tasks, such as text-to-video or layout-to-image generation, where a single conditioning embedding currently has to serve both realism and semantic fidelity.
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 / 6 minor

Summary. The manuscript proposes DTE-GAN, an end-to-end text-to-image GAN that learns two separate Bi-LSTM text embeddings: a generator-side embedding optimized through the adversarial/generation losses and a discriminator-side embedding optimized through a multi-modal contrastive loss on real image-text pairs. The generator conditions on both embeddings, while the discriminator uses the discriminator-side embedding for contrastive alignment and for adversarial criticism. The paper reports IS, FID, and R-precision on CUB, Oxford-102, and MS-COCO, presents an internal ablation (Table 6) showing that the dual-embedding configuration outperforms shared-embedding variants, demonstrates integration into AttnGAN (Table 4), and shows reuse of the learned embeddings for text-guided image manipulation (Table 5).

Significance. If the central claim holds, decoupling a text representation optimized for photo-realism from one optimized for text-image alignment is a useful, parameter-efficient design principle for lightweight GAN-based text-to-image synthesis. The paper has real strengths: Table 6 is a controlled internal ablation that isolates the embedding organization, Table 4 tests portability to an existing architecture, Table 5 shows downstream reuse, and DTE-GAN uses far fewer parameters (11M) than DF-GAN and SSA-GAN. The main reservation is that R-precision is computed with the same discriminator-side encoder that the generator is trained to align with, so the alignment benefit attributed to the dual-embedding design is not independently confirmed. An external-encoder evaluation would substantially strengthen the core claim.

major comments (4)
  1. [Appendix C.4; Eq. (6), Eq. (11)] The manuscript computes R-precision using text features from the D-side Bi-LSTM sentence encoder and image features from the discriminator network (Appendix C.4). These are exactly the encoder and image feature extractor optimized by the contrastive losses in Eq. (6) and Eq. (11), and the generator is trained to maximize similarity of generated images to SD under this same encoder. Consequently, the R-precision numbers in Tables 1, 3, 4, and 6 partly measure whether the generator conditions on the evaluation encoder rather than generic text-image alignment. The CUB jump from 63.79 to 86.64 between Table 6 row 3 and row 5 illustrates this dependence. Please re-evaluate R-precision (or at least the main comparisons) with an external frozen text/image encoder, such as CLIP or a DAMSM encoder not trained inside the DTE-GAN pipeline, and report both sets of numbers. This is necessary to support the abstract's and conclusion's claims about text-to-image alignment.
  2. [Section 1; Section 4.2; Table 1] The abstract and Section 4.2 state that DTE-GAN reduces CUB FID 'from 14.06 to 13.67', but no method in Table 1 has a FID baseline of 14.06: TIME is 14.30, DF-GAN is 14.81, and DAE-GAN is 15.19. Please identify the correct baseline or correct the number, because as written the improvement claim is not verifiable from the reported tables.
  3. [Section 4.2; Table 1; Table 6] The paper attributes part of its cross-method superiority to the dual-embedding design, but the comparisons in Tables 1 and 2 involve models with different generators, discriminators, loss terms, and parameter counts; the manuscript itself notes that DTE-GAN halves the width of DF-GAN/SSA-GAN blocks. Table 6 varies only the embedding organization within DTE-GAN's own architecture, so it cannot separate the effect of the embedding scheme from the effect of the overall architecture in the Table 1 and Table 2 comparisons. Please either add architecture-matched baseline models with shared embeddings for the external comparisons or qualify the attribution accordingly.
  4. [Section 4.2; Table 1] The sentence 'On MS-COCO, we achieve similar performance of DF-GAN and SSA-GAN with fewer parameters' is not supported by Table 1 unless it refers to DTE-GAN+MAGP: DTE-GAN itself reports COCO FID 25.17, while DF-GAN reports 19.32 and SSA-GAN reports 19.37; DTE-GAN+MAGP reports 19.69. Please correct the sentence or explicitly state that the comparable result is achieved by DTE-GAN+MAGP.
minor comments (6)
  1. [Title page/author line] The author line reads 'Y eruru Asrar Ahmed'; this appears to be a formatting artifact and should be corrected to 'Yeruru Asrar Ahmed'.
  2. [Throughout] There are numerous typos and garbled words, including 'covaraince', 'regualarisation', 'apporach', 'geandator', 'Futher', and 'perpal' in Figure 4; the manuscript needs a careful proofreading pass.
  3. [Table 3] The CLIP and CLIP+Gemb baselines are not fully specified: it should be stated explicitly whether they use the same DTE-GAN generator/discriminator architecture or a different GAN architecture, since otherwise the comparison is difficult to interpret.
  4. [Table 5] The method name 'MANIGAN' should be 'ManiGAN' for consistency with the referenced work.
  5. [Section 4.3.6/Table 8 caption] The caption 'Single embeddings with outLG training achieves superior performance' is grammatically unclear; it should likely read 'with LG training' or 'without LG training' as appropriate.
  6. [Table 2; Table 3; Table 6] Table 2 states that R-precision scores are not available in the literature for Oxford-102, yet Tables 3 and 6 report Oxford R-precision values for DTE and CLIP variants; this is not contradictory, but the manuscript should clarify that the literature comparison lacks such numbers while the paper independently reports them.

Circularity Check

1 steps flagged · score 6.0 of 10

R-precision is measured with the same D-side encoder that the generator is trained to match, so the text-image alignment claim is partly circular; FID/IS remain externally grounded.

  1. fitted input called prediction [Appendix C.4; Eqs. (6)-(7), Sec. 3.2]
    "For R-precision, we obtain text features from D-side Bi-LSTM sentence encoder and image features from discriminator network. ... L^G_cont(ˆfvi,SDi) = − log exp(Sim(ˆfvi,SDi)) / ∑Nj=1 exp(Sim(ˆfvi,SDj)) ... Sim(fv,SD) = cos(fv,SD)/τ"

    R-precision is not an independent probe of text-image alignment: its text encoder is the D-side Bi-LSTM and its image encoder is the discriminator network, i.e., the exact feature pair coupled by the contrastive losses. The generator is trained with L^G_cont to maximize cos(ˆfv, SD)/τ for the caption, so retrieval with the same pair measures how well the model optimizes its own objective. Table 6's R-precision jump from 63.79 to 86.64 when SD→G is switched on is therefore partly a metric artifact: the generator now conditions on the very SD whose encoder is used for scoring. The claim that the second embedding improves alignment is not independently established; only the FID/IS photo-realism gains escape this circularity.

full rationale

The paper contains no load-bearing self-citations and its architecture derivation is self-contained. The central issue is evaluative: the R-precision metric is computed with the discriminator-side encoder that is co-trained with the generator through the multi-modal contrastive loss, so the text-image alignment results reduce, in part, to the training objective by construction. The image-quality claims, however, rely on Inception-v3 based FID and IS, which are external and independently meaningful. Because the main photorealism benefit is externally grounded but the alignment benefit is partly self-referential, a partial circularity score of 6 is appropriate.

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

The paper is an empirical architecture paper, not a derivation. The named free parameters are training choices that affect the reported numbers but are not fitted to a specific target. The axioms are standard domain assumptions for GAN-based text-to-image evaluation. No new physical or conceptual entities are introduced.

free parameters (4)
  • contrastive temperature tau = unspecified
    Used in Sim() in Eq. (7) and Eq. (11); no value is given in the paper, so the contrastive loss scale is undefined.
  • loss weights lambda1, lambda2, lambda3 = 1, 1, 1
    Set to 1 without ablation; the balance between reconstruction, contrastive, and adversarial terms could affect the dual-embedding advantage.
  • EMA decay = 0.999
    Inference uses exponential moving average weights with decay 0.999; the impact on final metrics is not analyzed.
  • truncation parameter = unspecified
    The generator uses the truncation trick but no truncation value is given, which can strongly affect FID and IS.
assumptions (4)
  • domain assumption Multi-modal contrastive loss between real image-text pairs trains a text embedding that captures text-image alignment.
    This is the standard DAMSM assumption from Xu et al. (2018), reused here for the discriminator-side encoder.
  • domain assumption FID and IS computed with the fixed Inception-v3 network are valid proxies for generated image quality.
    The paper relies on these external metrics without questioning their validity for the datasets used.
  • domain assumption The separation of gradients (generator-side embeddings receive only generator losses; discriminator-side embeddings receive only contrastive loss) is implemented correctly at training time.
    The paper states the stop-gradient/detach operations but provides no code to verify the exact gradient routing.
  • domain assumption Reported baselines from prior papers use the same evaluation protocol and are directly comparable.
    The tables aggregate numbers from previous publications; any differences in preprocessing, image resolution, or metric computation would affect comparisons.

how reviews work

0 comments
Cite this review

Pith. "Pith review of End-to-end Training for Text-to-Image Synthesis using Dual-Text Embeddings." pith.science (2026). https://pith.science/paper/KMX6ULIB

@misc{pith2026250201507,
  author       = {Pith},
  title        = {Pith review of: End-to-end Training for Text-to-Image Synthesis using Dual-Text Embeddings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KMX6ULIB}},
  note         = {Machine review of arXiv:2502.01507}
}
read the original abstract

Text-to-Image (T2I) synthesis is a challenging task that requires modeling complex interactions between two modalities ( i.e., text and image). A common framework adopted in recent state-of-the-art approaches to achieving such multimodal interactions is to bootstrap the learning process with pre-trained image-aligned text embeddings trained using contrastive loss. Furthermore, these embeddings are typically trained generically and reused across various synthesis models. In contrast, we explore an approach to learning text embeddings specifically tailored to the T2I synthesis network, trained in an end-to-end fashion. Further, we combine generative and contrastive training and use two embeddings, one optimized to enhance the photo-realism of the generated images, and the other seeking to capture text-to-image alignment. A comprehensive set of experiments on three text-to-image benchmark datasets (Oxford-102, Caltech-UCSD, and MS-COCO) reveal that having two separate embeddings gives better results than using a shared one and that such an approach performs favourably in comparison with methods that use text representations from a pre-trained text encoder trained using a discriminative approach. Finally, we demonstrate that such learned embeddings can be used in other contexts as well, such as text-to-image manipulation.

Figures

Figures reproduced from arXiv: 2502.01507 by the authors.

Figure 1
Figure 1. (a) Text and image are projected into a shared embedding space to enhance mutual informa [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of DTE-GAN architecture. DTE-GAN consists of three core components: i) a single-stage [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visual comparision of the images generated by DF-GAN (Tao et al., 2022) and DTE-GAN on [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Illustration of the images generated by HDGAN (Zhang et al., 2018) and those of DTE-GAN on [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Images generated using CLIP, CLIP + Learnable Generator side embeddings (CLIP + [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Examples of manipulated images generated by LightWeight GAN (Li et al., 2020b) using DTE [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: UpBlock used in Generator of DTE-GAN. zϵR 100 ∼ N (0, I) , SG ϵ R 256 , SD ϵ R 256 , WG ϵ R 256 , WD ϵ R 256 Linear(512) −→ 512 Conditional Augmentation(512) −→ 200 Linear(200+100) −→ (8 ∗ ch) × 4 × 4 UpBlock −→ (8 ∗ ch) × 8 × 8 UpBlock −→ (8 ∗ ch) × 16 × 16 UpBlock −→…
Figure 8
Figure 8. Figure 8: DownBlock used in Discriminator of DTE-GAN. [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 15 canonical work pages

  1. [2]

    Jun Cheng, Fuxiang Wu, Yanling Tian, Lei Wang, and Dapeng Tao

    URL https://arxiv.org/pd f/1810.01365.pdf. Jun Cheng, Fuxiang Wu, Yanling Tian, Lei Wang, and Dapeng Tao. Rifegan: Rich feature generation for text-to-image synthesis from prior knowledge. In2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10908–10917,

    arXiv

  2. [5]

    13 Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman

    URLhttps://arxiv.org/abs/2012.09841. 13 Oran Gafni, Adam Polyak, Oron Ashual, Shelly Sheynin, Devi Parikh, and Yaniv Taigman. Make-a-scene: Scene-based text-to-image generation with human priors,

    arXiv 2012

  3. [6]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In Yoshua Bengio and Yann LeCun (eds.),3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings,

    work page 2015

  4. [8]

    Bowen Li, Xiaojuan Qi, Thomas Lukasiewicz, and Philip H.S. Torr. Manigan: Text-guided image manipula- tion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020a. 14 Bowen Li, Xiaojuan Qi, Philip Torr, and Thomas Lukasiewicz. Lightweight generative adversarial networks for text-guided image manipulation. In...

    work page 2020

  5. [10]

    Maria-Elena Nilsback and Andrew Zisserman

    URLhttps: //arxiv.org/abs/2102.09672. Maria-Elena Nilsback and Andrew Zisserman. Automated flower classification over a large number of classes. In Indian Conference on Computer Vision, Graphics and Image Processing, Dec

    arXiv

  6. [11]

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever

    URLhttps://arxiv.org/abs/2103.00020. Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation.CoRR, abs/2102.12092,

    arXiv

  7. [12]

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen

    URLhttps://arxiv.org/ abs/2102.12092. Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents,

    arXiv

  8. [13]

    Ali Razavi, Aaron van den Oord, and Oriol Vinyals

    URLhttps://arxiv.org/abs/2204.06125. Ali Razavi, Aaron van den Oord, and Oriol Vinyals. Generating diverse high-fidelity images with vq-vae-2. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett (eds.),Advances in Neural Information Processing Systems. Curran Associates, Inc.,

    arXiv

Show all 27 references
  1. [14]

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen

    URL https://arxiv.org/abs/2205.11487. Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. InProceedings of the 30th International Conference on Neural Information Processing Systems, NIPS’16, pp. 2234–...

    arXiv

  2. [16]

    Hongchen Tan, Xiuping Liu, Baocai Yin, and Xin Li

    doi: 10.1109/TIP.2020.3026728. Hongchen Tan, Xiuping Liu, Baocai Yin, and Xin Li. Cross-modal semantic matching generative adversarial networks for text-to-image synthesis.IEEE Transactions on Multimedia, 24:832–845, 2022a. doi: 10.110 9/TMM.2021.3060291. Hongchen Tan, Xiuping...

    2020

  3. [17]

    Ming Tao, Hao Tang, Fei Wu, Xiao-Yuan Jing, Bing-Kun Bao, and Changsheng Xu

    doi: 10.1109/ICIBA50161 .2020.9277299. Ming Tao, Hao Tang, Fei Wu, Xiao-Yuan Jing, Bing-Kun Bao, and Changsheng Xu. Df-gan: A simple and effective baseline for text-to-image synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), ...

    2020

  4. [21]

    Stackgan++: Realistic image synthesis with stacked generative adversarial networks.IEEE Transactions on Pattern Analysis and Machine Intelligence, PP, 10 2017a

    Han Zhang, Tao Xu, Hongsheng Li, Shaoting Zhang, Xiaogang Wang, Xiaolei Huang, and Dimitris Metaxas. Stackgan++: Realistic image synthesis with stacked generative adversarial networks.IEEE Transactions on Pattern Analysis and Machine Intelligence, PP, 10 2017a. doi: 10.1109/TP...

  5. [23]

    Zizhao Zhang, Yuanpu Xie, and Lin Yang

    URLhttps://arxiv.org/abs/2011.02709. Zizhao Zhang, Yuanpu Xie, and Lin Yang. Photographic text-to-image synthesis with a hierarchically-nested adversarial network. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June

    2011 arXiv

  6. [25]

    A DTE-GAN with MA-GP Recent methods (Tao et al., 2022; Zhang & Schomaker, 2020; Liao et al.,

    2022

  7. [26]

    of replicating sentence features and concatenating with image features to predict logit values for adversarial loss, the modified adversarial loss function of Discriminator for conditional loss with MA-GP is: LAdv D =− Ex∼Pdata[max(0, 1−D(x,SD))] + Eˆx∼PG[max(0, 1 +D(ˆx,SD))] ...

    1997

  8. [27]

    (2017) in both the generator and discriminator text encoders, and we have reported the results in Table

    with a 4-layer Transformer encoder Vaswani et al. (2017) in both the generator and discriminator text encoders, and we have reported the results in Table

    2017

  9. [28]

    DTE-GAN architecture consists of a dual text embedding setup (Section C.1), a single-stage Generator (Section C.2) and a Discriminator (Section C.3)

    framework. DTE-GAN architecture consists of a dual text embedding setup (Section C.1), a single-stage Generator (Section C.2) and a Discriminator (Section C.3). C.1 Dual Text Embeddings IntheDualText Embeddingssetup, bi-DirectionalLSTM(Schuster&Paliwal,1997)areusedastextencode...

    1997

  10. [29]

    The generatorG takes noisez along with generator-side sentence embeddingsSG and the discriminator-side sentence embeddingSD and passes them through a set of linear layers followed by a set of upsampling blocks (UpBlocks). UpBlock at each stage is utilised for up sampling spati...

    2019

  11. [200]

    Technical Report CNS-TR-2010-001, California Institute of Technology,

    2010

  12. [1997]

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli

    doi: 10.1109/78.650093. Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Francis Bach and David Blei (eds.),Proceedings of the 32nd International Conference on Machine Learning, volum...

    doi

  13. [2015]

    Alex Krizhevsky

    URLhttp://arxiv.org/abs/1412.6980. Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report,

    arXiv

  14. [2017]

    URLhttps://proceedings.neurips.cc/paper/2017/file /3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf. P. Welinder, S. Branson, T. Mita, C. Wah, F. Schroff, S. Belongie, and P. Perona. Caltech-UCSD Birds

    2017

  15. [2018]

    Lafite: Towards language-free training for text-to-image generation.arXiv preprint arXiv:2111.13792,

    Yufan Zhou, Ruiyi Zhang, Changyou Chen, Chunyuan Li, Chris Tensmeyer, Tong Yu, Jiuxiang Gu, Jinhui Xu, and Tong Sun. Lafite: Towards language-free training for text-to-image generation.arXiv preprint arXiv:2111.13792,

    arXiv

  16. [2019]

    Ting Chen, Mario Lučić, Neil Houlsby, and Sylvain Gelly

    doi: 10.1109/ACCESS.2019.2958864. Ting Chen, Mario Lučić, Neil Houlsby, and Sylvain Gelly. On self-modulation for generative adversarial networks. InInternational Conference on Learning Representations,

    2019

  17. [2020]

    Ayushman Dash, John Cristian Borges Gamboa, Sheraz Ahmed, Marcus Liwicki, and Muhammad Zeshan Afzal

    doi: 10.1109/CVPR42600.2020.01092. Ayushman Dash, John Cristian Borges Gamboa, Sheraz Ahmed, Marcus Liwicki, and Muhammad Zeshan Afzal. Tac-gan - text conditioned auxiliary classifier generative adversarial network,

    2020

  18. [2022]

    Lawrence Zitnick

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. Microsoft coco: Common objects in context. In David Fleet, Tomas Pajdla, Bernt Schiele, and Tinne Tuytelaars (eds.),Computer Vision – ECCV 2014, 2014a. ...

    2014

  19. [2024]

    17 Guojun Yin, Bin Liu, Lu Sheng, Nenghai Yu, Xiaogang Wang, and Jing Shao

    doi: 10.1109/TMM.2023.3266607. 17 Guojun Yin, Bin Liu, Lu Sheng, Nenghai Yu, Xiaogang Wang, and Jing Shao. Semantics disentangling for text-to-image generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June

    2023

Pith tools

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