pith. sign in

arxiv: 2606.23950 · v1 · pith:ZZXFHG2Ynew · submitted 2026-06-22 · 💻 cs.CV

DivRL: Disentangled Self-Similarity Rewards for Diverse Subject-Driven Generation

Qian Wang , Zhenyu Li , Abdelrahman Eldesokey , Peter Wonka This is my paper

Pith reviewed 2026-06-26 08:35 UTC · model grok-4.3

classification 💻 cs.CV
keywords subject-driven generationstructural diversityidentity preservationself-similarity measuregated optimizationimage generationreinforcement learning
0
0 comments X

The pith

Treating identity as a gated constraint lets subject-driven generators explore structural diversity without losing subject consistency.

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

The paper targets the identity-diversity paradox in subject-driven image generation, where preserving a subject's appearance tends to produce rigid outputs with little variation. It introduces DivRL as a post-training method that defines Negative Self-Similarity Measure to score structural diversity and Visual Semantic Matching to score identity consistency. The central mechanism is an Explore-and-Suppress strategy that lets the model sample freely diverse structures and applies a quadratic hinge penalty only when identity falls below a threshold. This formulation turns identity preservation from a competing objective into a feasibility check, so both measures can rise together. A reader would care because the approach yields subject-driven outputs that remain recognizable while showing greater structural variety.

Core claim

DivRL jointly optimizes identity consistency and structural diversity by using disentangled features from a similarity model; Negative Self-Similarity Measure quantifies diversity while Visual Semantic Matching quantifies identity, and the Explore-and-Suppress strategy treats the latter as a gated constraint that penalizes only threshold violations via quadratic hinge loss, converting identity preservation into a feasibility constraint rather than a competing term.

What carries the argument

The Explore-and-Suppress strategy, which treats Visual Semantic Matching as a gated constraint so the model can freely sample configurations scored high by Negative Self-Similarity Measure and penalizes only those that violate the identity threshold.

If this is right

  • The generator produces images with higher structural diversity at comparable identity consistency levels.
  • Identity preservation no longer trades off directly against diversity because it functions only as a post-exploration filter.
  • Post-training with the gated loss improves both objectives simultaneously rather than requiring separate balancing.
  • The method applies to existing subject-driven pipelines without retraining the base generator from scratch.

Where Pith is reading between the lines

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

  • The same gated-constraint pattern could be tested on other conflicting objectives such as text alignment versus visual realism.
  • If the disentanglement holds, the approach might reduce the need for heavy prompt engineering in creative workflows.
  • A practical test would measure whether users prefer the outputs when asked to generate multiple distinct views of the same subject.

Load-bearing premise

A similarity model already supplies visual features disentangled enough that one measure tracks only structural diversity and the other tracks only identity.

What would settle it

An experiment in which applying the identity gate causes measured structural diversity to stop increasing or causes identity consistency to fall below the baseline model.

Figures

Figures reproduced from arXiv: 2606.23950 by Abdelrahman Eldesokey, Peter Wonka, Qian Wang, Zhenyu Li.

Figure 1
Figure 1. Figure 1: Our method achieves a better balance between structural diversity, prompt following, and identity consistency compared with strong baselines. Abstract. Subject-driven image generation faces an “Identity-Diversity Paradox”, where strong identity preservation often leads to rigid and low￾diversity outputs. We propose a post-training framework called DivRL that jointly optimizes identity consistency and struc… view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline of our method DivRL. We design reward models VSM for identity consistency (IC) and nSSM for structural diversity (SD). A two-stage “Explore-and￾Suppress” strategy is used to generate images that are both identity-consistent and structurally diverse. The first stage encourages free exploration for diverse generation, while the second stage employs a gater to filter out low-consistency samples, main… view at source ↗
Figure 3
Figure 3. Figure 3: Quantitative evaluation of the trade-off between identity consistency and struc￾tural diversity. Results are reported using both in-domain metrics (MTG) and out-of￾domain metrics (DINO). We further analyze the trade-off between consistency and diversity in Fig￾ure 3 using the consistency ratio and diversity-over-consistency metrics. Both in￾domain and out-of-domain evaluations show that non-Flux-Kontext-ba… view at source ↗
Figure 4
Figure 4. Figure 4: The examples include both rigid objects ( [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison with baselines. Our method produces structurally diverse images while preserving identity consistency and maintaining good prompt alignment. times fail to follow the prompt interactions. This behavior stems from the strong emphasis on identity preservation, which encourages the model to reproduce a subject configuration similar to the reference rather than adapting it to the contextual in… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison between optimizing using nSSM only, VSM only, and ours. We highlight cases of identity inconsistency, prompt misalignment, and struc￾tural rigidity. Our method DivRL achieves the best balance between prompt following, identity preservation, and structural diversity. 4.4 Ablation Studies Single-Reward Optimization We further compare our method with the two reward variants in [PITH_FU… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison between our optimization strategy and linear reward weighting. Left is the training progression curve, while right is the trade-off frontier between the consistency ratio and diversity-over-consistency. For the linear weighting annotation, it is denoted as the linear ratio between VSM and nSSM. Reference Gen 1 Gen 2 Image Mask SSM map nSSM 0.199 0.331 -- [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual￾ization of the SSM maps and the correspond￾ing nSSM scores. Original feature Original feature Original feature Downsampled feature Downsampled feature Downsampled feature [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison between the original Flux-Kontext and DivRL under the multi￾prompt setting. E.2 Multi-seed comparison In [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Comparison between the original Flux-Kontext and DivRL under the multi￾seed setting. ated by Flux-Kontext exhibit identity drift across different seeds, whereas our approach maintains consistent subject identity. E.3 Comparison with more baselines We provide an additional visual comparison between our method and other baseline models in [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison between multiple baseline models and our method DivRL. In contrast, when the structure (pose) of the main foreground object varies while the overall semantics say the same, the nSSM score has a sharper increase. nSSM score peaks when both structural and semantic changes exist. Therefore, despite being able to respond to both semantic and structural changes, nSSM is more sensitive to structural … view at source ↗
Figure 12
Figure 12. Figure 12: How nSSM responds to semantic/structural changes. Flux-Kontext Ours Flux-Kontext Ours “… sitting at a Cafe” “… reading a newspaper” [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Failure case: both the original Flux-Kontext model and our method may produce high-frequency noises of the identity region on the generated images. filter to mitigate the amplification of these artifacts during RL, we leave further architectural refinements for pixel-level visual fidelity to future work. Additionally, despite its robustness, our method occasionally struggles to achieve an optimal balance … view at source ↗
Figure 14
Figure 14. Figure 14: Failure case: our method occasionally exhibits a trade-off between prompt adherence and identity consistency [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
read the original abstract

Subject-driven image generation faces an "Identity-Diversity Paradox", where strong identity preservation often leads to rigid and low-diversity outputs. We propose a post-training framework called DivRL that jointly optimizes identity consistency and structural diversity simultaneously by leveraging disentangled visual features from a robust similarity model. Specifically, we introduce a Negative Self-Similarity Measure (nSSM) to quantify structural diversity, and Visual Semantic Matching (VSM) to evaluate identity consistency. We propose an "Explore-and-Suppress" strategy that treats VSM as a gated constraint: the model freely explores structurally diverse configurations, and only samples that violate the identity threshold are penalized via a quadratic hinge loss. This converts identity preservation from a competing objective into a feasibility constraint, allowing nSSM and VSM to improve jointly. Experiments demonstrate that our method effectively pushes the model to generate both consistent and diverse images and improves structural diversity while maintaining comparable identity consistency through a gated optimization formulation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper proposes DivRL, a post-training framework for subject-driven image generation to resolve the Identity-Diversity Paradox. It introduces nSSM to quantify structural diversity and VSM to measure identity consistency, using an Explore-and-Suppress strategy that treats VSM as a gated feasibility constraint (with quadratic hinge loss on violations) so the model can explore diverse configurations while preserving identity, allowing joint improvement via disentangled features from a robust similarity model.

Significance. If the disentanglement assumption holds and the gated optimization produces verifiable joint gains, the approach could offer a practical way to convert competing objectives into a constraint in generative modeling, with potential impact on subject-driven synthesis tasks. The paper ships no machine-checked proofs, reproducible code, or parameter-free derivations, so significance rests entirely on the empirical claims.

major comments (3)
  1. [Abstract] Abstract: the statement that 'experiments demonstrate that our method effectively pushes the model to generate both consistent and diverse images' supplies no quantitative metrics, baselines, ablation tables, or statistical comparisons, so the central claim of joint nSSM/VSM improvement rests on an unverified assertion.
  2. [Abstract] Abstract (and method description): nSSM and VSM are defined within the paper and then used both to drive the Explore-and-Suppress training objective and to evaluate success; without external benchmarks, held-out similarity models, or shipped code, it is impossible to determine whether reported gains are independent of the measure definitions themselves.
  3. [Abstract] Abstract (Explore-and-Suppress paragraph): the strategy assumes the underlying similarity model already supplies features with negligible residual entanglement between structure (nSSM) and identity (VSM), yet no quantitative check (feature orthogonality, mutual information, or ablation on entangled vs. disentangled backbones) is described; if correlation remains, the VSM gate will either suppress valid diversity or fail to protect identity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment below with references to the full manuscript and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the statement that 'experiments demonstrate that our method effectively pushes the model to generate both consistent and diverse images' supplies no quantitative metrics, baselines, ablation tables, or statistical comparisons, so the central claim of joint nSSM/VSM improvement rests on an unverified assertion.

    Authors: We agree that the abstract would be strengthened by including key quantitative results. The full manuscript (Section 4 and associated tables) reports concrete metrics, baseline comparisons (e.g., DreamBooth, LoRA), and ablation studies demonstrating joint gains in nSSM and VSM. We will revise the abstract to summarize these specific improvements and statistical comparisons. revision: yes

  2. Referee: [Abstract] Abstract (and method description): nSSM and VSM are defined within the paper and then used both to drive the Explore-and-Suppress training objective and to evaluate success; without external benchmarks, held-out similarity models, or shipped code, it is impossible to determine whether reported gains are independent of the measure definitions themselves.

    Authors: nSSM and VSM are inherently task-specific, so their use in both the objective and evaluation is by design. The manuscript mitigates circularity through comparisons against multiple baselines on standard subject-driven benchmarks and sensitivity analyses across similarity model variants. We acknowledge that the current submission does not include shipped code or additional held-out models; we will add a limitations discussion on this point and commit to releasing code upon acceptance to support independent verification. revision: partial

  3. Referee: [Abstract] Abstract (Explore-and-Suppress paragraph): the strategy assumes the underlying similarity model already supplies features with negligible residual entanglement between structure (nSSM) and identity (VSM), yet no quantitative check (feature orthogonality, mutual information, or ablation on entangled vs. disentangled backbones) is described; if correlation remains, the VSM gate will either suppress valid diversity or fail to protect identity.

    Authors: The similarity model was chosen based on prior evidence of disentangled representations. While explicit orthogonality or mutual-information analyses were not included in the initial submission, the empirical joint gains observed support practical effectiveness of the gating. We will add a dedicated analysis subsection with feature correlation metrics and backbone ablations in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained.

full rationale

The paper introduces nSSM and VSM as new measures derived from an external robust similarity model, then defines an Explore-and-Suppress training strategy that uses VSM as a constraint while optimizing nSSM. No equations or self-citations are shown that reduce the claimed joint improvement to a tautology or fitted input by construction. The disentanglement assumption is stated but not derived from the measures themselves; the framework remains an independent proposal whose empirical validity would be checked against held-out evaluations or human studies rather than reducing to its own definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; all such elements remain unknown.

pith-pipeline@v0.9.1-grok · 5698 in / 1090 out tokens · 20822 ms · 2026-06-26T08:35:19.170779+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

44 extracted references · 15 canonical work pages · 10 internal anchors

  1. [1]

    Black, K., Janner, M., Du, Y., Kostrikov, I., Levine, S.: Training diffusion models with reinforcement learning (2023)

    2023

  2. [2]

    BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset

    Chen, J., Xu, Z., Pan, X., Hu, Y., Qin, C., Goldstein, T., Huang, L., Zhou, T., Xie, S., Savarese, S., et al.: Blip3-o: A family of fully open unified multimodal models-architecture, training and dataset. arXiv preprint arXiv:2505.09568 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Chen, X., Huang, L., Liu, Y., Shen, Y., Zhao, D., Zhao, H.: Anydoor: Zero-shot object-level image customization. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 6593–6602 (2024)

    2024

  4. [4]

    In: Advances in Neural Information Processing Systems (2025)

    Eldesokey, A., Cvejic, A., Ghanem, B., Wonka, P.: Mind-the-glitch: Visual corre- spondence for detecting inconsistencies in subject-driven generation. In: Advances in Neural Information Processing Systems (2025)

    2025

  5. [5]

    In: Forty-first international conference on machine learning (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: Forty-first international conference on machine learning (2024)

    2024

  6. [6]

    Advances in Neural Information Processing Sys- tems36, 79858–79885 (2023)

    Fan, Y., Watkins, O., Du, Y., Liu, H., Ryu, M., Boutilier, C., Abbeel, P., Ghavamzadeh, M., Lee, K., Lee, K.: Dpok: Reinforcement learning for fine-tuning text-to-image diffusion models. Advances in Neural Information Processing Sys- tems36, 79858–79885 (2023)

    2023

  7. [7]

    Gal, R., Alaluf, Y., Atzmon, Y., Patashnik, O., Bermano, A.H., Chechik, G., Cohen-Or, D.: An image is worth one word: Personalizing text-to-image genera- tionusingtextualinversion.In:TheEleventhInternationalConferenceonLearning Representations (2023),https://openreview.net/forum?id=NAQvF08TcyG

    2023

  8. [8]

    In: The Thirty-ninth Annual Conference on Neu- ral Information Processing Systems (2025),https://openreview.net/forum?id= cZMno8E3yp

    Goyal, A., Qian, G., Coskun, H., Gupta, A., Tam, H., Ostashev, D., Hu, J., Sagar, D., Tulyakov, S., Aberman, K., Wang, K.C.: Preventing shortcuts in adapter train- ing via providing the shortcuts. In: The Thirty-ninth Annual Conference on Neu- ral Information Processing Systems (2025),https://openreview.net/forum?id= cZMno8E3yp

    2025

  9. [9]

    In: The Fourteenth International Conference on Learning Representations (2026),https://openreview.net/forum?id=7mCo3R3Wyn

    He, X., Fu, S., Zhao, Y., Li, W., Yang, J., Yin, D., Rao, F., Zhang, B.: TEMPFLOW-GRPO:WHENTIMINGMATTERSFORGRPOINFLOWMOD- ELS. In: The Fourteenth International Conference on Learning Representations (2026),https://openreview.net/forum?id=7mCo3R3Wyn

    2026

  10. [10]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Hu, Z., Zhang, F., Chen, L., Kuang, K., Li, J., Gao, K., Xiao, J., Wang, X., Zhu, W.: Towards better alignment: Training diffusion models with reinforcement learn- ing against sparse rewards. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 23604–23614 (2025)

    2025

  11. [11]

    Advances in neural information processing systems36, 36652–36663 (2023)

    Kirstain,Y.,Polyak,A.,Singer,U.,Matiana,S.,Penna,J.,Levy,O.:Pick-a-pic:An open dataset of user preferences for text-to-image generation. Advances in neural information processing systems36, 36652–36663 (2023)

    2023

  12. [12]

    In: IEEE International Conference on Computer Vision (ICCV) (2025)

    Kumari, N., Yin, X., Zhu, J.Y., Misra, I., Azadi, S.: Generating multi-image syn- thetic data for text-to-image customization. In: IEEE International Conference on Computer Vision (ICCV) (2025)

    2025

  13. [13]

    FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space

    Labs,B.F.,Batifol,S.,Blattmann,A.,Boesel,F.,Consul,S.,Diagne,C.,Dockhorn, T., English, J., English, Z., Esser, P., Kulal, S., Lacey, K., Levi, Y., Li, C., Lorenz, D., Müller, J., Podell, D., Rombach, R., Saini, H., Sauer, A., Smith, L.: Flux.1 kontext: Flow matching for in-context image generation and editing in latent space (2025),https://arxiv.org/abs/2...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  14. [14]

    In: Thirty-seventh Conference on Neural Information Processing Systems (2023),https://openreview.net/forum? id=g6We1SwaY9

    Li, D., Li, J., Hoi, S.: BLIP-diffusion: Pre-trained subject representation for con- trollable text-to-image generation and editing. In: Thirty-seventh Conference on Neural Information Processing Systems (2023),https://openreview.net/forum? id=g6We1SwaY9

    2023

  15. [15]

    MixGRPO: Unlocking Flow-based GRPO Efficiency with Mixed ODE-SDE

    Li, J., Cui, Y., Huang, T., Ma, Y., Fan, C., Yang, M., Zhong, Z.: Mixgrpo: Unlock- ing flow-based grpo efficiency with mixed ode-sde. arXiv preprint arXiv:2507.21802 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  16. [16]

    In: European Conference on Computer Vision

    Lin, Z., Pathak, D., Li, B., Li, J., Xia, X., Neubig, G., Zhang, P., Ramanan, D.: Evaluating text-to-visual generation with image-to-text generation. In: European Conference on Computer Vision. pp. 366–384. Springer (2024)

    2024

  17. [17]

    In: The Thirty- ninthAnnualConferenceonNeuralInformationProcessingSystems(2025),https: //openreview.net/forum?id=oCBKGw5HNf

    Liu, J., Liu, G., Liang, J., Li, Y., Liu, J., Wang, X., Wan, P., ZHANG, D., Ouyang, W.: Flow-GRPO: Training flow matching models via online RL. In: The Thirty- ninthAnnualConferenceonNeuralInformationProcessingSystems(2025),https: //openreview.net/forum?id=oCBKGw5HNf

    2025

  18. [18]

    arXiv preprint arXiv:2512.02014 (2025)

    Liu, Z., Ren, W., Liu, H., Zhou, Z., Chen, S., Qiu, H., Huang, X., An, Z., Yang, F., Patel, A., et al.: Tuna: Taming unified visual representations for native unified multimodal models. arXiv preprint arXiv:2512.02014 (2025)

    work page arXiv 2025

  19. [19]

    SpatialReward: Bridging the Perception Gap in Online RL for Image Editing via Explicit Spatial Reasoning

    Long, Y., Yang, Y., Wei, H., Chen, W., Zhang, T., Liu, C., Jiang, K., Chen, J., Tang, K., Wen, B., et al.: Spatialreward: Bridging the perception gap in online rl for image editing via explicit spatial reasoning. arXiv preprint arXiv:2602.07458 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    Editscore: Unlocking online rl for image editing via high-fidelity reward modeling, 2026

    Luo, X., Wang, J., Wu, C., Xiao, S., Jiang, X., Lian, D., Zhang, J., Liu, D., et al.: Editscore: Unlocking online rl for image editing via high-fidelity reward modeling. arXiv preprint arXiv:2509.23909 (2025)

    work page arXiv 2025

  21. [21]

    arXiv preprint arXiv:2510.14256 (2025)

    Meng, X., Zhang, Z., Zhang, Z., Liao, J., Qin, L., Wang, W.: Identity-grpo: Opti- mizing multi-human identity-preserving video generation via reinforcement learn- ing. arXiv preprint arXiv:2510.14256 (2025)

    work page arXiv 2025

  22. [22]

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

    Miao, Z., Wang, J., Wang, Z., Yang, Z., Wang, L., Qiu, Q., Liu, Z.: Training diffusion models towards diverse image generation with reinforcement learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 10844–10853 (2024)

    2024

  23. [23]

    In: Proceedings of the Computer Vision and Pattern Recog- nition Conference

    Na, S., Kim, Y., Lee, H.: Boost your human image generation model via direct pref- erence optimization. In: Proceedings of the Computer Vision and Pattern Recog- nition Conference. pp. 23551–23562 (2025)

    2025

  24. [24]

    Advances in neural information processing sys- tems35, 27730–27744 (2022)

    Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al.: Training language models to follow instructions with human feedback. Advances in neural information processing sys- tems35, 27730–27744 (2022)

    2022

  25. [25]

    In: The Thirteenth International Conference on Learning Representa- tions (2025),https://dreambenchplus.github.io/

    Peng, Y., Cui, Y., Tang, H., Qi, Z., Dong, R., Bai, J., Han, C., Ge, Z., Zhang, X., Xia, S.T.: Dreambench++: A human-aligned benchmark for personalized image generation. In: The Thirteenth International Conference on Learning Representa- tions (2025),https://dreambenchplus.github.io/

    2025

  26. [26]

    arXiv preprint arXiv:2512.04784 (2025)

    Ping, B., Jia, C., Luo, M., Xia, C., Shen, X., Dang, Z., Qian, H.: Paco-rl: Advanc- ing reinforcement learning for consistent image generation with pairwise reward modeling. arXiv preprint arXiv:2512.04784 (2025)

    work page arXiv 2025

  27. [27]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    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. pp. 10684–10695 (2022)

    2022

  28. [28]

    DivRL 17 In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2023)

    Ruiz, N., Li, Y., Jampani, V., Pritch, Y., Rubinstein, M., Aberman, K.: Dream- booth: Fine tuning text-to-image diffusion models for subject-driven generation. DivRL 17 In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2023)

    2023

  29. [29]

    In: SIGGRAPH Asia 2024 Conference Papers

    Sauer, A., Boesel, F., Dockhorn, T., Blattmann, A., Esser, P., Rombach, R.: Fast high-resolution image synthesis with latent adversarial diffusion distillation. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–11 (2024)

    2024

  30. [30]

    Shao, Z., Wang, P., Zhu, Q., Xu, R., Song, J., Bi, X., Zhang, H., Zhang, M., Li, Y.K., Wu, Y., Guo, D.: Deepseekmath: Pushing the limits of mathematical reasoning in open language models (2024),https://arxiv.org/abs/2402.03300

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  31. [31]

    In: Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition

    Shi, J., Xiong, W., Lin, Z., Jung, H.J.: Instantbooth: Personalized text-to-image generation without test-time finetuning. In: Proceedings of the IEEE/CVF confer- ence on computer vision and pattern recognition. pp. 8543–8552 (2024)

    2024

  32. [32]

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

    Wallace, B., Dang, M., Rafailov, R., Zhou, L., Lou, A., Purushwalkam, S., Ermon, S., Xiong, C., Joty, S., Naik, N.: Diffusion model alignment using direct preference optimization. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8228–8238 (June 2024)

    2024

  33. [33]

    Wu, C., Li, J., Zhou, J., Lin, J., Gao, K., Yan, K., ming Yin, S., Bai, S., Xu, X., Chen, Y., Chen, Y., Tang, Z., Zhang, Z., Wang, Z., Yang, A., Yu, B., Cheng, C., Liu, D., Li, D., Zhang, H., Meng, H., Wei, H., Ni, J., Chen, K., Cao, K., Peng, L., Qu, L., Wu, M., Wang, P., Yu, S., Wen, T., Feng, W., Xu, X., Wang, Y., Zhang, Y., Zhu, Y., Wu, Y., Cai, Y., L...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  34. [34]

    OmniGen2: Towards Instruction-Aligned Multimodal Generation

    Wu, C., Zheng, P., Yan, R., Xiao, S., Luo, X., Wang, Y., Li, W., Jiang, X., Liu, Y., Zhou, J., Liu, Z., Xia, Z., Li, C., Deng, H., Wang, J., Luo, K., Zhang, B., Lian, D., Wang, X., Wang, Z., Huang, T., Liu, Z.: Omnigen2: Exploration to advanced multimodal generation. arXiv preprint arXiv:2506.18871 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Editreward: A human-aligned reward model for instruction-guided image editing, 2026

    Wu, K., Jiang, S., Ku, M., Nie, P., Liu, M., Chen, W.: Editreward: A human-aligned reward model for instruction-guided image editing. arXiv preprint arXiv:2509.26346 (2025)

    work page arXiv 2025

  36. [36]

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

    Wu, S., Huang, M., Wu, W., Cheng, Y., Ding, F., He, Q.: Less-to-more general- ization: Unlocking more controllability by in-context generation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 18682–18692 (2025)

    2025

  37. [37]

    Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis

    Wu, X., Hao, Y., Sun, K., Chen, Y., Zhu, F., Zhao, R., Li, H.: Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis. arXiv preprint arXiv:2306.09341 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  38. [38]

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

    Xiao, S., Wang, Y., Zhou, J., Yuan, H., Xing, X., Yan, R., Li, C., Wang, S., Huang, T., Liu, Z.: Omnigen: Unified image generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 13294– 13304 (June 2025)

    2025

  39. [39]

    In: Proceedingsofthe37thInternationalConferenceonNeuralInformationProcessing Systems

    Xu, J., Liu, X., Wu, Y., Tong, Y., Li, Q., Ding, M., Tang, J., Dong, Y.: Imagere- ward: learning and evaluating human preferences for text-to-image generation. In: Proceedingsofthe37thInternationalConferenceonNeuralInformationProcessing Systems. pp. 15903–15935 (2023)

    2023

  40. [40]

    DanceGRPO: Unleashing GRPO on Visual Generation

    Xue, Z., Wu, J., Gao, Y., Kong, F., Zhu, L., Chen, M., Liu, Z., Liu, W., Guo, Q., Huang, W., et al.: Dancegrpo: Unleashing grpo on visual generation. arXiv preprint arXiv:2505.07818 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  41. [41]

    Ye, H., Zhang, J., Liu, S., Han, X., Yang, W.: Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models (2023)

    2023

  42. [42]

    Wang et al

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models (2023) 18 Q. Wang et al

    2023

  43. [43]

    Secrets of RLHF in Large Language Models Part I: PPO

    Zheng, R., Dou, S., Gao, S., Hua, Y., Shen, W., Wang, B., Liu, Y., Jin, S., Liu, Q., Zhou, Y., et al.: Secrets of rlhf in large language models part i: Ppo. arXiv preprint arXiv:2307.04964 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    motorcycle

    Zhu, H., Xiao, T., Honavar, V.G.: DSPO: Direct score preference optimization for diffusion model alignment. In: The Thirteenth International Conference on Learn- ing Representations (2025),https://openreview.net/forum?id=xyfb9HHvMe A Evaluation metrics A.1 Structural DINO (sDINO) sDINO measures the structural similarity between the patches across the refe...

    2025