Pith. sign in

REVIEW 4 major objections 7 minor 69 references

On the Fairness, Diversity and Reliability of Text-to-Image Generative Models

T0 review · 4 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a text-to-image model's sensitivity to small embedding perturbations reveals its reliability, and that the resulting reliability, diversity, and fairness scores can detect intentionally biased models and retrieve…

desk verdict A useful evaluation framework for T2I reliability, diversity, and fairness, but the detection/provenance claims outrun the experiments due to lacking benign fine-tuning controls and statistical grounding. read the letter →

arxiv 2411.13981 v2 pith:J6UYQ3SU submitted 2024-11-21 cs.CV cs.AI

Jordan Vice , Naveed Akhtar , Leonid Sigal , Richard Hartley , Ajmal Mian This is my paper · ORCID
classification cs.CVcs.AI
keywords text-to-imagegenerationmodelreliabilitygenerativefairnessdiversitybackdoordetectiontriggerretrievalembeddingperturbationslatentdiffusionmodels
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

The paper tries to establish that the trustworthiness of a text-to-image model can be read off from how its generated images react to small random perturbations of the encoded prompt. It packages this into global and local reliability scores, then adds two further scores: generative diversity, the breadth of images a concept produces, and generative fairness, how much control a single token exerts when removed from the prompt. Across three families of intentionally biased models and benign Stable Diffusion baselines, the biased models show measurably higher perturbation sensitivity, lower diversity on trigger concepts, and lower fairness for the trigger token. The significance is practical: a training-free audit can flag suspicious models and even name the token responsible for the bias.

What carries the argument

The central object is the embedding perturbation $\varphi_E$, a scaled random multiplicative vector applied to text-encoder embeddings: $\varphi_E^G = \delta_p \sigma_x$ for the whole prompt and $\varphi_E^L = \delta_p \sigma_{x_i}$ for a single token. The perturbation is increased stepwise until the cosine similarity between the original and perturbed image falls below $0.9$, and the perturbation value at that crossing is recorded. These recorded values form reliability distributions whose left shift and peak height signal unreliability. Diversity is the average pairwise dis-similarity of $N$ images from a single-token prompt, and fairness is $-\log(1 - \cos \theta)$ for a leave-one-out prompt under a low guidance scale.

What would settle it

Generate images for "drink" from the benign SD-V1.5 model under the paper's single-token, low-guidance protocol and compute its diversity and fairness scores; if they match or fall below the BAGM trigger values of around 0.150 and 2.115 reported in Table 3, the claim that these scores reveal intentional bias rather than concept specificity is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the core discovery is that reliability is not a single property but a distribution: global reliability measures how many prompts break under embedding perturbation, and local reliability measures which tokens within those prompts are responsible. The authors show that this separation matters, since the TPA rare-trigger model escapes global reliability testing while its trigger token is plainly visible in the local distribution. They then demonstrate that the diversity score identifies natural-language triggers such as "drink" and "burger", while the fairness score identifies the rare trigger "ˆo", establishing bias provenance. The paper states this as evidence that intentionally biased models are simultaneously unreliable and unfair, and that both characteristics are observable from outputs and text embeddings alone.

Load-bearing premise

The fairness and diversity stages only isolate injected bias if an unusually strong token influence and unusually homogeneous images cannot arise from an ordinary, semantically specific concept in a benign model.

Editorial extensions

If this is right

  • A grey-box audit of any open text-to-image model can flag specific prompts and tokens that drive unreliable behavior, without retraining.
  • Because the diversity and fairness stages need only prompts and images, they can also be run against black-box models.
  • The trigger-retrieval split is actionable: natural-language triggers surface through low diversity, and rare triggers surface through low fairness.
  • Intentionally biased models are detectable even when only 10% of test prompts contain the trigger.
  • The method is training-free and search-free, so it can serve as a first-pass check before heavier backdoor defenses are invoked.

Reading between the lines

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

  • We infer that the same perturbation-sensitivity idea could extend to text-to-video or text-to-audio generators, where small changes to condition embeddings may similarly expose injected behavior.
  • We infer that the low-guidance leave-one-out fairness score is an implicit measure of a token's causal influence on generation, and calibrating it against attribution methods could sharpen trigger retrieval.
  • We infer a testable rule from the ontology ablation: a concept whose measured diversity falls below its parent class by more than the paper's observed drops of roughly 0.068 to 0.288 points may be a candidate for intentional bias, though the paper itself does not propose such a threshold.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes an evaluation framework for text-to-image (T2I) models based on perturbations applied to text-encoder embeddings. It defines global reliability RG and local reliability RL as distributions of the perturbation magnitude needed to change a generated image below a cosine-similarity threshold (τφE = 0.9), and then feeds tokens identified as unreliable into two further stages: generative diversity D~xT, computed from pairwise similarities among images generated from a single token, and generative fairness F~xT, computed from leave-one-out prompt ablations under low guidance. The method is validated on presumed-benign models (SD-V1.4/1.5/2.1) and intentionally-biased/backdoored models (BAGM, TPA, BadT2I), with claims that the framework detects biased models and retrieves bias triggers. The paper includes ablations over ViT encoders, prompt distributions, denoising steps, and concept hierarchies, and the code is publicly released.

Significance. If the causal confounds are addressed, this would be a useful and relatively inexpensive auditing tool for openly available T2I models: it requires only text-encoder outputs and generated images, and it attempts to characterize reliability, fairness, and diversity under a single framework. The strengths of the paper are its public code, the breadth of ablations, and the fact that the core reliability experiments use three independently constructed backdoor benchmarks. The left shifts in Table 2 and Fig. 6 for BAGM and BadT2I, and the local-reliability detection of the TPA trigger, are indicative but not yet conclusive. The main claims of bias detection and provenance retrieval are currently under-supported because the comparisons lack benign fine-tuning controls, statistical significance measures, and a quantitative trigger-retrieval criterion.

major comments (4)
  1. [§3.6, Table 2] The intentional-bias comparisons lack benign fine-tuned controls. BAGM, TPA, and BadT2I are all fine-tuned versions of SD-V1.5, and BAGM is fine-tuned on the narrow Marketable Foods dataset, yet the only baselines are off-the-shelf SD-V1.4/1.5/2.1. A benign SD-V1.5 model fine-tuned on the same MF data could plausibly exhibit a left-shifted PG/L(φE) and low D~xT for concepts such as 'drink' or 'coffee' simply because the training distribution is domain-specific, without any backdoor manipulation. This confounding of fine-tuning/dataset shift with bias injection directly affects the central detection claim. Please add control experiments with benign models fine-tuned on the same data, or otherwise demonstrate that the observed shifts are not produced by fine-tuning alone.
  2. [§3.3, §3.4, Table 3] The trigger-retrieval analysis does not compare the same concept across benign and biased models. For example, 'drink' is reported with D~xT=0.150 in the BAGM column, but there is no 'drink' entry for benign SD-V1.5, and the ontology ablation in Table 6 shows that specificity naturally lowers diversity. Without a within-concept comparison, a low diversity value for a BAGM trigger cannot be attributed to the injected bias rather than to the semantic specificity of the token or to the fine-tuning dataset. The same issue applies to the F~xT analysis. Please report D~xT and F~xT for an identical set of prompts across all models, including benign fine-tuned controls.
  3. [§3.2, Table 2, Fig. 6] There are no error bars, confidence intervals, or significance tests anywhere in the reliability comparisons. The perturbations use a random vector ℜ, and the paper does not state the number of independent runs or the values of N, Nptb, and NK used in the main experiments. Some reported differences are small; for example, SD-V1.5 has RG φMo=0.1233 while BadT2I has φMo=0.1155, which may be within run-to-run noise given the stochastic generation process. Please provide variance estimates, multiple seeds, or bootstrap intervals to substantiate the distribution-shift claims.
  4. [§3.4, §4, Table 3] The claim that F~xT retrieves the TPA rare trigger is not supported by the reported table. In the TPA F~xT column of Table 3, 'ˆo' has F~xT=1.519, whereas 'plate' has F~xT=1.060, 'poster' has 1.400, and 'clock' has 1.495, meaning those tokens show a greater (not smaller) influence on generation under the authors' own definition. The paper does not define a quantitative retrieval criterion (e.g., a rank threshold, z-score, or comparison to a benign baseline), and Fig. 8 is qualitative. Please define the retrieval protocol and evaluate it with a metric such as hit@k or ROC over the full token set.
minor comments (7)
  1. [§3.4, Eq. (7)] Equation (7) is missing norm notation in the denominator; it should read ||I~xTk|| ||I~x||. Please also state the numerical value of the low guidance scale used for F~xT evaluations, since 'low guidance' is never quantified.
  2. [§3.2, Eq. (4)] The expression '1−φE ≤ ℜ ≤ 1+φE' and the scaling '1+φE / 1−φE' are ambiguous without parentheses; please clarify the intended interval and the division.
  3. [Table 3] The header 'R ~xT' and the special rows 'X' and 'N→N-2' are not self-explanatory; a legend or a short caption explaining how the rows are ordered and which rows correspond to triggers would greatly improve readability.
  4. [§4, Table 4] The Conceptual Captions ablation in Table 4 evaluates only benign models, so it cannot verify that the benign-vs-biased reliability contrast persists under a different input distribution; please include at least one intentionally-biased model in this ablation.
  5. [§4, ViT comparison] The ViT ablation states that a 'smaller test prompt set' was used, but the sample size is not reported; please state it so the reader can judge the strength of the consistency claim.
  6. [§3.5] The sentence 'our method is training-free and search-free' should be qualified: the reliability stage requires repeated generation over a potentially large perturbation grid, and the computational cost of this is acknowledged in the Limitations but not quantified.
  7. [Throughout] There are several typos and formatting artifacts, including 'T able' in table captions, 'All models use the the AdamW optimizer' in Table 1, and 'a persˆon' in Fig. 5. These should be corrected in a final pass.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: reliability, diversity, and fairness metrics are defined directly from image/embedding similarities, and the self-cited BAGM attack serves as an independently reproduced validation benchmark rather than a fitted premise.

full rationale

The paper's derivation chain is self-contained. RG and RL are computed from cosine similarity between original and perturbed-image embeddings (Eqs. 3-5), D~xT from pairwise similarity over N generated images (Eq. 6), and F~xT from leave-one-out similarity under low guidance (Eq. 7). None of these quantities is fitted to the labels 'biased' or to the trigger tokens of BAGM, TPA, or BadT2I; they are operational measurements defined before any benchmark is introduced. The BAGM benchmark [16] is prior work by the same authors, but it is code-reproduced from provided code and its trigger/target pairs are defined by the published attack; moreover, the framework is also validated on independent attacks TPA [1] and BadT2I [17], so the self-citation is not load-bearing. The ontology ablation (Figs. 11-12, Table 6) provides an expectation heuristic for concept specificity, but it does not enter the metric definitions and is not used to fit any threshold in Eqs. 3-7. The absence of a benign-fine-tuned SD-V1.5 control is a real threat to causal attribution of left-shifts to bias injection, but that is an experimental confound and thus a correctness risk, not a circularity. Accordingly, no circular step can be exhibited, and the score is 0.

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

The framework's central metrics rest on several unproven modeling assumptions: on-manifold behavior of embedding perturbations, validity of CLIP cosine similarity for semantic change, low-guidance leave-one-out as a fairness measure, and generated-image diversity as a proxy for training-data diversity. These are not free parameters fitted to the target result, but they are load-bearing and should be validated with sensitivity analysis and same-concept controls.

free parameters (4)
  • delta_p (perturbation step size) = 0.05 in the example; final experimental value not reported
    Chosen by hand in Section 3.2 Eq. (3); determines which prompts are flagged as globally unreliable.
  • tau_phiE (similarity threshold) = 0.9
    Set in Section 3.2 to avoid drastic scene changes; it defines unreliable behavior and receives no sensitivity analysis.
  • Guidance scale for fairness evaluations = not numerically specified
    Section 3.4 states a significantly reduced guidance scale is used; the exact value is omitted, making the fairness results hard to reproduce precisely.
  • Number of generated images N, NK, Nptb = not specified
    Used in Eq. (5), Eq. (6), and Eq. (7); the text only says N random images, so the sample sizes for the reported statistics are unclear.
assumptions (4)
  • domain assumption Embedding perturbations of the form x * R with R bounded by 1 +/- phiE stay on the learned manifold for small delta_p.
    Stated in footnote 2 of Section 3.2; off-manifold perturbations would conflate image change with unreliability.
  • domain assumption Cosine similarity in CLIP ViT image space is a valid measure of semantic alignment between generated images.
    Used in Eqs. (5)-(7) and all evaluations; no independent validation that CLIP similarity tracks human-judged semantic change for these concepts.
  • domain assumption Removing a token and measuring output change under low guidance isolates that token's influence on generation.
    Section 3.4; if low guidance still confounds syntactic or semantic role, F~xT measures general importance rather than fairness.
  • domain assumption Generated-image diversity for a single-token prompt reflects the diversity of the learned concept.
    Section 3.3; relies on training data diversity propagating to output diversity, which is plausible but not proven for arbitrary concepts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the Fairness, Diversity and Reliability of Text-to-Image Generative Models." pith.science (2026). https://pith.science/paper/J6UYQ3SU

@misc{pith2026241113981,
  author       = {Pith},
  title        = {Pith review of: On the Fairness, Diversity and Reliability of Text-to-Image Generative Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6UYQ3SU}},
  note         = {Machine review of arXiv:2411.13981}
}
read the original abstract

The rapid proliferation of multimodal generative models has sparked critical discussions on their reliability, fairness and potential for misuse. While text-to-image models excel at producing high-fidelity, user-guided content, they often exhibit unpredictable behaviors and vulnerabilities that can be exploited to manipulate class or concept representations. To address this, we propose an evaluation framework to assess model reliability by analyzing responses to global and local perturbations in the embedding space, enabling the identification of inputs that trigger unreliable or biased behavior. Beyond social implications, fairness and diversity are fundamental to defining robust and trustworthy model behavior. Our approach offers deeper insights into these essential aspects by evaluating: (i) generative diversity, measuring the breadth of visual representations for learned concepts, and (ii) generative fairness, which examines the impact that removing concepts from input prompts has on control, under a low guidance setup. Beyond these evaluations, our method lays the groundwork for detecting unreliable, bias-injected models and tracing the provenance of embedded biases. Our code is publicly available at https://github.com/JJ-Vice/T2I_Fairness_Diversity_Reliability. Keywords: Fairness, Reliability, AI Ethics, Bias, Text-to-Image Models

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

69 extracted references · 47 canonical work pages

  1. [1]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Struppek, L., Hintersdorf, D., Kersting, K.: Rickrolling the artist: Injecting back- doors into text encoders for text-to-image synthesis. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 4584–4596 (2023)

    work page 2023

  2. [2]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV), pp

    D’Inc` a, M., Tzelepis, C., Patras, I., Sebe, N.: Improving fairness using vision- language driven image augmentation. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV), pp. 4695–4704 (2024)

    work page 2024

  3. [3]

    AI and Ethics, 1–21 (2024)

    Friedrich, F., Brack, M., Struppek, L., Hintersdorf, D., Schramowski, P., Luccioni, S., Kersting, K.: Auditing and instructing text-to-image generation models on fairness. AI and Ethics, 1–21 (2024)

    work page 2024

  4. [4]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Luo, Y., Shi, M., Khan, M.O., Afzal, M.M., Huang, H., Yuan, S., Tian, Y., Song, L., Kouhana, A., Elze, T., Fang, Y., Wang, M.: Fairclip: Harnessing fairness in vision-language learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 12289–12301 (2024)

    work page 2024

  5. [5]

    Artificial Intelligence Review 57(322) (2024) https://doi.org/10.1007/ s10462-024-10974-1

    Kashima, H., Oyama, S., Arai, H., et al.: Trustworthy human computation: a survey. Artificial Intelligence Review 57(322) (2024) https://doi.org/10.1007/ s10462-024-10974-1

    work page 2024

  6. [6]

    Artificial Intelligence Review 58(35) (2025) https://doi.org/10.1007/ s10462-024-11040-6 23

    Roth, B., Araujo, P.H., Xia, Y., et al.: Specification overfitting in artificial intelligence. Artificial Intelligence Review 58(35) (2025) https://doi.org/10.1007/ s10462-024-11040-6 23

    work page 2025

  7. [7]

    Artificial Intelligence Review 57(243) (2024) https://doi.org/10.1007/s10462-024-10896-y

    Lin, Z., Guan, S., Zhang, W., et al.: Towards trustworthy llms: a review on debi- asing and dehallucinating in large language models. Artificial Intelligence Review 57(243) (2024) https://doi.org/10.1007/s10462-024-10896-y

    doi:10.1007/s10462-024-10896-y 2024

  8. [8]

    Artificial Intelligence Review 58(99) (2025) https: //doi.org/10.1007/s10462-025-11110-3

    Chen, H., Xiang, Q., Hu, J., et al.: Comprehensive exploration of diffusion models in image generation: a survey. Artificial Intelligence Review 58(99) (2025) https: //doi.org/10.1007/s10462-025-11110-3

    doi:10.1007/s10462-025-11110-3 2025

Show all 69 references
  1. [9]

    ACM Computing Surveys 54(6), 1–35 (2021)

    Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., Galstyan, A.: A survey on bias and fairness in machine learning. ACM Computing Surveys 54(6), 1–35 (2021)

    2021

  2. [10]

    In: Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society

    Abid, A., Farooqi, M., Zou, J.: Persistent anti-muslim bias in large language models. In: Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society. AIES ’21, pp. 298–306 (2021). https://doi.org/10.1145/3461702.3462624 . https://doi.org/10.1145/3461702.3462624

    2021

  3. [11]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp

    Cho, J., Zala, A., Bansal, M.: Dall-eval: Probing the reasoning skills and social biases of text-to-image generation models. In: Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 3043–3054 (2023)

    2023

  4. [12]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Feng, Y., Shah, C.: Has ceo gender bias really been fixed? adversarial attacking and improving gender fairness in image search. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, pp. 11882–11890 (2022)

    2022

  5. [13]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    D’Inc` a, M., Peruzzo, E., Mancini, M., Xu, D., Goel, V., Xu, X., Wang, Z., Shi, H., Sebe, N.: Openbias: Open-set bias detection in text-to-image generative models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 12225–12235 (2024)

    2024

  6. [14]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Bai, J., Gao, K., Min, S., Xia, S.-T., Li, Z., Liu, W.: Badclip: Trigger-aware prompt learning for backdoor attacks on clip. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 24239– 24250 (2024)

    2024

  7. [15]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Huang, Y., Juefei-Xu, F., Guo, Q., Zhang, J., Wu, Y., Hu, M., Li, T., Pu, G., Liu, Y.: Personalization as a shortcut for few-shot backdoor attack against text- to-image diffusion models. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 21169–21178 (2024)

    2024

  8. [16]

    IEEE Transactions on Information Forensics and Security, 1–1 (2024) https://doi.org/10.1109/TIFS.2024.3386058

    Vice, J., Akhtar, N., Hartley, R., Mian, A.: Bagm: A backdoor attack for manipulating text-to-image generative models. IEEE Transactions on Information Forensics and Security, 1–1 (2024) https://doi.org/10.1109/TIFS.2024.3386058

    2024

  9. [17]

    In: Proceed- ings of the 31st ACM International Conference on Multimedia

    Zhai, S., Dong, Y., Shen, Q., Pu, S., Fang, Y., Su, H.: Text-to-image diffusion 24 models can be easily backdoored through multimodal data poisoning. In: Proceed- ings of the 31st ACM International Conference on Multimedia. MM ’23, pp. 1577–

  10. [18]

    https://arxiv.org/abs/2311

    Shen, X., Du, C., Pang, T., Lin, M., Wong, Y., Kankanhalli, M.: Finetuning Text-to-Image Diffusion Models for Fairness (2024). https://arxiv.org/abs/2311. 07604

    2024

  11. [19]

    Advances in Neural Information Processing Systems 36 (2024)

    Teo, C., Abdollahzadeh, M., Cheung, N.-M.M.: On measuring fairness in gener- ative models. Advances in Neural Information Processing Systems 36 (2024)

    2024

  12. [20]

    International Journal of Human–Computer Interaction 36(6), 495–504 (2020) https://doi.org/10.1080/10447318.2020.1741118

    Shneiderman, B.: Human-centered artificial intelligence: Reliable, safe & trust- worthy. International Journal of Human–Computer Interaction 36(6), 495–504 (2020) https://doi.org/10.1080/10447318.2020.1741118

    2020

  13. [21]

    Science and Engineering Ethics 26(5), 2749–2767 (2020)

    Ryan, M.: In ai we trust: ethics, artificial intelligence, and reliability. Science and Engineering Ethics 26(5), 2749–2767 (2020)

    2020

  14. [22]

    Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and Harnessing Adversarial Examples (2015)

    2015

  15. [23]

    arXiv (2019) 1706.06083 [stat.ML]

    Madry, A., Makelov, A., Schmidt, L., Tsipras, D., Vladu, A.: Towards deep learn- ing models resistant to adversarial attacks. arXiv (2019) 1706.06083 [stat.ML]

    2019 arXiv

  16. [24]

    Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., Fergus, R.: Intriguing properties of neural networks (2014)

    2014

  17. [25]

    IEEE Access 6, 14410–14430 (2018) https://doi.org/10.1109/ ACCESS.2018.2807385

    Akhtar, N., Mian, A.: Threat of adversarial attacks on deep learning in computer vision: A survey. IEEE Access 6, 14410–14430 (2018) https://doi.org/10.1109/ ACCESS.2018.2807385

    2018

  18. [26]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017)

    Moosavi-Dezfooli, S.-M., Fawzi, A., Fawzi, O., Frossard, P.: Universal adversarial perturbations. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017)

    2017

  19. [27]

    IEEE Access 9, 155161–155196 (2021) https://doi.org/10.1109/ACCESS.2021.3127960

    Akhtar, N., Mian, A., Kardan, N., Shah, M.: Advances in adversarial attacks and defenses in computer vision: A survey. IEEE Access 9, 155161–155196 (2021) https://doi.org/10.1109/ACCESS.2021.3127960

    2021

  20. [28]

    arXiv preprint arXiv:1312.6114 (2013)

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

    2013 arXiv

  21. [29]

    In: Advances in Neural Information Processing Systems (NeurIPS), pp

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial nets. In: Advances in Neural Information Processing Systems (NeurIPS), pp. 2672–2680 (2014) 25

    2014

  22. [30]

    In: International Confer- ence on Machine Learning, pp

    Sohl-Dickstein, J., Weiss, E., Maheswaranathan, N., Ganguli, S.: Deep unsuper- vised learning using nonequilibrium thermodynamics. In: International Confer- ence on Machine Learning, pp. 2256–2265 (2015). PMLR

    2015

  23. [31]

    Advances in neural information processing systems 33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems 33, 6840–6851 (2020)

    2020

  24. [32]

    arXiv preprint arXiv:2010.02502 (2020)

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502 (2020)

    2020 arXiv

  25. [33]

    arXiv preprint arXiv:2011.13456 (2020)

    Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 (2020)

    2020 arXiv

  26. [34]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10684– 10695 (2022)

    2022

  27. [35]

    arXiv preprint arXiv:2204.06125 (2022) [cs.CV]

    Ramesh, A., Dhariwal, P., Nichol, A., Chu, C., Chen, M.: Hierarchical text- conditional image generation with clip latents. arXiv preprint arXiv:2204.06125 (2022) [cs.CV]

    2022 arXiv

  28. [36]

    : Photorealistic text-to-image diffusion models with deep language understanding

    Saharia, C., Chan, W., Saxena, S., et al. : Photorealistic text-to-image diffusion models with deep language understanding. In: Advances in Neural Information Processing Systems, vol. 35, pp. 36479–36494 (2022)

    2022

  29. [37]

    Advances in neural information processing systems 30 (2017)

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. Advances in neural information processing systems 30 (2017)

    2017

  30. [38]

    : Learning transferable visual models from natural language supervision

    Radford, A., Kim, J.W., Hallacy, C., et al. : Learning transferable visual models from natural language supervision. In: Meila, M., Zhang, T. (eds.) Proceedings of the 38th International Conference on Machine Learning. Proceedings of Machine Learning Research, vol. 139, pp. 87...

    2021

  31. [39]

    Philippand Brox: U-net: Convolutional networks for biomedical image segmentation

    Ronneberger, O., Fischer, T. Philippand Brox: U-net: Convolutional networks for biomedical image segmentation. In: Navab, N., Hornegger, J., Wells, W.M., Frangi, A.F. (eds.) Medical Image Computing and Computer-Assisted Interven- tion – MICCAI 2015, pp. 234–241. Springer, Cham (2015)

    2015

  32. [40]

    ArXiv (2024) arXiv:2403.03431 [cs.CV]

    Liu, B., Wang, C., Cao, T., Jia, K., Huang, J.: Towards understanding cross and self-attention in stable diffusion for text-guided image editing. ArXiv (2024) arXiv:2403.03431 [cs.CV]

    2024 arXiv

  33. [41]

    arXiv 26 preprint arXiv:2301.12247 (2023)

    Brack, M., Friedrich, F., Hintersdorf, D., Struppek, L., Schramowski, P., Kerst- ing, K.: Sega: Instructing text-to-image models using semantic guidance. arXiv 26 preprint arXiv:2301.12247 (2023)

    2023 arXiv

  34. [42]

    In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Guo, J., Xu, X., Pu, Y., Ni, Z., Wang, C., Vasu, M., Song, S., Huang, G., Shi, H.: Smooth diffusion: Crafting smooth latent spaces in diffusion models. In: Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7548–7558 (2024)

    2024

  35. [43]

    In: Proceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision (W ACV), pp

    Bhatt, G., Balasubramanian, V.N.: Learning style subspaces for controllable unpaired domain translation. In: Proceedings of the IEEE/CVF Winter Confer- ence on Applications of Computer Vision (W ACV), pp. 4220–4229 (2023)

    2023

  36. [44]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Shi, Y., Yang, X., Wan, Y., Shen, X.: Semanticstylegan: Learning compositional generative priors for controllable image synthesis and editing. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 11254–11264 (2022)

    2022

  37. [45]

    In: Advances in Neural Information Processing Systems, vol

    Luccioni, S., Akiki, C., Mitchell, M., Jernite, Y.: Stable bias: Evaluating soci- etal representations in diffusion models. In: Advances in Neural Information Processing Systems, vol. 36, pp. 56338–56351 (2023)

    2023

  38. [46]

    https://arxiv.org/abs/ 2405.17814

    Luo, H., Deng, Z., Chen, R., Liu, Z.: F AIntbench: A Holistic and Precise Bench- mark for Bias Evaluation in Text-to-Image Models (2024). https://arxiv.org/abs/ 2405.17814

    2024 arXiv

  39. [47]

    In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

    Chinchure, A., Shukla, P., Bhatt, G., Salij, K., Hosanagar, K., Sigal, L., Turk, M.: Tibet: Identifying and evaluating biases in text-to-image generative models. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024,...

    2024

  40. [48]

    arXiv preprint arXiv:2312.13053 (2023)

    Vice, J., Akhtar, N., Hartley, R., Mian, A.: Quantifying bias in text-to-image generative models. arXiv preprint arXiv:2312.13053 (2023)

    2023 arXiv

  41. [49]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Bakr, E.M., Sun, P., Shen, X., Khan, F.F., Li, L.E., Elhoseiny, M.: Hrs-bench: Holistic, reliable and scalable benchmark for text-to-image models. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 20041–20053 (2023)

    2023

  42. [50]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp

    Hu, Y., Liu, B., Kasai, J., Wang, Y., Ostendorf, M., Krishna, R., Smith, N.A.: Tifa: Accurate and interpretable text-to-image faithfulness evaluation with ques- tion answering. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 20406–20417 (2023)

    2023

  43. [51]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Chen, W., Song, D., Li, B.: Trojdiff: Trojan attacks on diffusion models with diverse targets. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4035–4044 (2023)

    2023

  44. [52]

    4015–4024 (2023)

    Chou, S.-Y., Chen, P.-Y., Ho, T.-Y.: How to backdoor diffusion models? In: 27 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4015–4024 (2023)

    2023

  45. [53]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    An, S., Chou, S.-Y., Zhang, K., Xu, Q., Tao, G., Shen, G., Cheng, S., Ma, S., Chen, P.-Y., Ho, T.-Y., et al.: Elijah: Eliminating backdoors injected in diffusion models via distribution shift. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 1084...

    2024

  46. [54]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Zhu, L., Ning, R., Li, J., Xin, C., Wu, H.: Seer: Backdoor detection for vision- language models through searching target text and image trigger jointly. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, pp. 7766–7774 (2024)

    2024

  47. [55]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Feng, S., Tao, G., Cheng, S., Shen, G., Xu, X., Liu, Y., Zhang, K., Ma, S., Zhang, X.: Detecting backdoors in pre-trained encoders. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 16352–16362 (2023)

    2023

  48. [56]

    In: ECCV (2024)

    Wang, Z., Zhang, J., Shan, S., Chen, X.: T2ishield: Defending against backdoors on text-to-image diffusion models. In: ECCV (2024)

    2024

  49. [57]

    https://arxiv.org/abs/2404

    Guan, Z., Hu, M., Li, S., Vullikanti, A.: UFID: A Unified Framework for Input- level Backdoor Detection on Diffusion Models (2024). https://arxiv.org/abs/2404. 01101

    2024

  50. [58]

    https://huggingface.co/runwayml/ stable-diffusion-v1-5 (2023)

    RunwayML: runwayml/stable-diffusion-v1-5. https://huggingface.co/runwayml/ stable-diffusion-v1-5 (2023)

    2023

  51. [59]

    https://huggingface.co/ stable-diffusion-v1-5/stable-diffusion-v1-5 (2024)

    sd-legacy: stable-diffusion-v1-5/stable-diffusion-v1-5. https://huggingface.co/ stable-diffusion-v1-5/stable-diffusion-v1-5 (2024)

    2024

  52. [60]

    https://huggingface.co/CompVis/ stable-diffusion-v1-4 (2023)

    CompVis: CompVis/stable-diffusion-v1-4. https://huggingface.co/CompVis/ stable-diffusion-v1-4 (2023)

    2023

  53. [61]

    https://huggingface.co/stabilityai/ stable-diffusion-2-1 (2024)

    AI, S.: stabilityai/stable-diffusion-2-1. https://huggingface.co/stabilityai/ stable-diffusion-2-1 (2024)

    2024

  54. [62]

    https://ieee-dataport.org/documents/marketable-foods-mf-dataset (2023)

    Vice, J., Akhtar, N., Hartley, R., Mian, A.: Marketable Foods (MF) Dataset. https://ieee-dataport.org/documents/marketable-foods-mf-dataset (2023)

    2023

  55. [63]

    In: Fleet, D., Pajdla, T., Schiele, B., Tuytelaars, T

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll´ ar, P., Zitnick, C.L.: Microsoft coco: Common objects in context. In: Fleet, D., Pajdla, T., Schiele, B., Tuytelaars, T. (eds.) Proceedings of the European Conference on Computer Vision, pp. 740–755 (2014)

    2014

  56. [64]

    In: Proceed- ings of the 56th Annual Meeting of the Association for Computational Linguistics, pp

    Sharma, P., Ding, N., Goodman, S., Soricut, R.: Conceptual captions: A cleaned, 28 hypernymed, image alt-text dataset for automatic image captioning. In: Proceed- ings of the 56th Annual Meeting of the Association for Computational Linguistics, pp. 2556–2565 (2018)

    2018

  57. [65]

    Wu, B., Xu, C., Dai, X., Wan, A., Zhang, P., Yan, Z., Tomizuka, M., Gonzalez, J., Keutzer, K., Vajda, P.: Visual Transformers: Token-based Image Representation and Processing for Computer Vision (2020)

    2020

  58. [66]

    In: Meila, M., Zhang, T

    Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., Jegou, H.: Training data-efficient image transformers and distillation through attention. In: Meila, M., Zhang, T. (eds.) Proceedings of the 38th International Conference on Machine Learning. Proceedings of Machine...

    2021

  59. [67]

    Pattern Recognition Letters 105, 63–70 (2018)

    Choi, J., Lee, K., Torralba, A.: Visual concept ontology for image annotations. Pattern Recognition Letters 105, 63–70 (2018)

    2018

  60. [68]

    In: NeurIPS (2021) 29

    Zhao, Y., Andreas, J., Parikh, D., Mottaghi, R.: A benchmark for systematic generalization in grounded language understanding. In: NeurIPS (2021) 29

    2021

  61. [1587]

    https: //doi.org/10.1145/3581783.3612108

    Association for Computing Machinery, New York, NY, USA (2023). https: //doi.org/10.1145/3581783.3612108 . https://doi.org/10.1145/3581783.3612108

    2023

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.