Pith. sign in

REVIEW 4 major objections 5 minor 6 cited by

MOSAIC claims that explicitly supervising point-to-point attention correspondence between references and target, plus an orthogonal-attention disentanglement loss, keeps identity fidelity in multi-subject generation with four or more refere

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

MOSAIC improves multi-subject personalized image generation by supervising attention maps with semantic point correspondences and a disentanglement loss, and introduces the SemAlign-MS dataset for training.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Genuinely new dataset idea and sensible losses, but the 4+ subject claim has no quantitative support and the correspondence algorithm is never described. the 4 major comments →

arxiv 2509.01977 v1 pith:2O7IYVZM submitted 2025-09-02 cs.CV

MOSAIC: Multi-Subject Personalized Generation via Correspondence-Aware Alignment and Disentanglement

Dong She , Siming Fu , Mushui Liu , Qiaoqiao Jin , Hualiang Wang , Mu Liu , Jidong Jiang This is my paper
classification cs.CV
keywords multi-subject personalizationtext-to-image generationsemantic correspondenceattention supervisiondisentanglementdiffusion transformersubject-driven generation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

MOSAIC claims that multi-subject personalized image generation fails when attention is never told which reference region should map to which target region, and when different subjects' representations are allowed to interfere. To fix this, the authors build a new dataset of 1.2 million image pairs with dense semantic point correspondences between reference subjects and target images, then train a diffusion transformer with two additional losses: one that supervises reference-to-target attention at those annotated points, and one that pushes different references' attention patterns apart. On the DreamBench and XVerseBench benchmarks, they report the strongest identity-fidelity scores among the compared methods, and they report that fidelity holds even with four or more reference subjects, a regime where existing methods visibly degrade.

Core claim

The paper's central claim is that multi-subject generation should be treated as a representation-alignment problem: the model needs to know exactly which target-latent positions each reference token should attend to, and different subjects should occupy separate attention subspaces. With SemAlign-MS, a 1.2M-pair dataset of annotated reference-to-target point correspondences, MOSAIC trains a diffusion-transformer generator with the semantic correspondence attention loss (LSCA) and the multi-reference disentanglement loss (LMD). LSCA applies cross-entropy to the averaged reference-to-target attention at each annotated point pair, forcing each reference token to focus on its designated target l

What carries the argument

The central object is the reference-to-target attention sub-matrix A_ref→tgt inside the diffusion transformer's multi-modal attention, averaged across all DiT blocks. LSCA supervises this matrix with cross-entropy at annotated point pairs (u,v), forcing each reference token to attend to its corresponding target-latent token. LMD aggregates per-reference attention vectors, normalizes them, and maximizes their symmetric KL divergence, pushing subjects into orthogonal attention subspaces. A correspondence-disjointness constraint ensures every target token is assigned to at most one reference, preventing ambiguous supervision where different subjects compete for the same region.

Load-bearing premise

The method assumes that the automatically established semantic point correspondences between reference and target images are accurate enough to supervise attention, and that supervising attention toward a visible target token during training transfers to generation, where the target token does not yet exist.

What would settle it

A controlled experiment that replaces the SemAlign-MS correspondences with random point pairs during training—keeping the same losses and data—would test whether the specific correspondence labels are load-bearing; if multi-subject CLIP-I and DINO drop little, the claimed mechanism is not what carries the result. A second check is to measure attention-peak accuracy at inference: whether the reference tokens actually concentrate on the target regions described in the prompt, rather than on unrelated parts of the generated latent.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the reported results hold, multi-subject personalized generation can scale beyond three references without identity blending, enabling group scenes with four or more distinct subjects from a handful of reference images.
  • SemAlign-MS supplies a training signal—dense point correspondences—that was previously unavailable, making explicit attention-level supervision possible for the first time in this setting.
  • The two losses are additive to a base diffusion-transformer objective, suggesting the same supervision can be dropped into other reference-conditioned DiT generators.
  • The correspondence-disjointness rule implies target regions can be cleanly partitioned among references, directly addressing attribute leakage at the token level.
  • MOSAIC reports gains in both single-subject and multi-subject settings, meaning the representation-level alignment also helps the simpler one-reference case, not only scenes with many subjects.

Where Pith is reading between the lines

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

  • Beyond the paper: the same attention-shaping recipe could be applied at inference time using predicted correspondences extracted from diffusion features, removing the need for a labelled dataset.
  • Beyond the paper: if attention divergence is the mechanism behind 4+ subject stability, then prompting-time interventions that reweight attention maps to enforce separation could reproduce part of the effect without retraining.
  • Beyond the paper: the reported train/inference gap—the target latent exists during training but not at generation—suggests the gains could partly come from the model learning to localize subjects in the noisy latent; a control with random correspondences would reveal how much of the gain depends on the specific point labels.
  • Beyond the paper: the orthogonal-subspace idea likely extends beyond images to any multi-entity generation task, such as multi-person video or multi-object 3D scenes, where the same interference problem arises.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes MOSAIC, a training-time framework for multi-subject personalized generation built on FLUX-1.0-dev with a LoRA branch for reference-image processing. The method introduces SemAlign-MS, a synthetically constructed dataset of 1.2M image pairs with claimed semantic point correspondences between reference subjects and target images, plus two new losses: a semantic correspondence attention loss (LSCA) that supervises reference-to-target attention at correspondence points via cross-entropy, and a multi-reference disentanglement loss (LMD) that maximizes symmetric KL divergence between different references' aggregate attention patterns. The total loss is L = L_diff + αLSCA + βLMD with α=0.4, β=0.6. Evaluations on DreamBench and XVerseBench report improved single- and multi-subject metrics over existing baselines, and qualitative examples illustrate generation with up to five reference subjects. Ablations in Table 3 show that each loss improves the multi-subject metrics.

Significance. If the claims hold, the work is significant: it is one of the first attempts to explicitly supervise semantic point correspondences in multi-subject personalization, and the attention-level alignment/disentanglement formulation is simple and conceptually clean. The SemAlign-MS dataset, if released with the correspondence metadata, could be a useful community resource. The method is not circular: the evaluation benchmarks are external and do not set the loss weights, and the ablations give indirect evidence that both losses contribute. However, the headline differentiator—sustained fidelity with 4+ reference subjects—is not quantitatively supported anywhere in the paper, and Section 3 leaves the correspondence construction underspecified at precisely the step that the two losses depend on. These issues must be resolved before the central claims can be accepted.

major comments (4)
  1. [Abstract; Section 5.2.1, Tables 1 and 2; Fig. 4 row 6] The central claim that MOSAIC 'maintains high fidelity with 4+ reference subjects' is not supported by any quantitative per-subject-count evaluation. Tables 1 and 2 report only aggregate single-subject vs. multi-subject scores; the subject-count distribution of the multi-subject splits is not given, and no per-K metrics (K=2,3,4,5,6) are reported. The only direct 4+ evidence is a single qualitative row (Fig. 4, row 6). With no breakdown, the aggregate multi-subject numbers could be driven by easier 2–3 subject cases, leaving the differentiating scalability claim unverified. Please report per-subject-count metrics with sample sizes and error bars, or remove/qualify the 4+ claim.
  2. [Section 3, Eqs. (1)–(2); Section 4.2, Eq. (10)] The load-bearing step in SemAlign-MS is the construction of semantic point correspondences, but the paper only states that 'we establish semantic point correspondences' (after Eq. 2) without describing the matching algorithm, the point-sampling strategy, the mapping from pixel coordinates to target-latent token positions, or any validation of correspondence accuracy. Because both LSCA and LMD are direct functions of C^(i,k), noisy or incorrect correspondences would propagate into both losses. Additionally, Eq. (2) asserts that the disjointness constraint V^(i,k1) ∩ V^(i,k2) = ∅ is satisfied for all training samples, but no evidence or algorithm for enforcing/certifying this is given. Please provide the concrete pipeline and a quantitative analysis (e.g., coverage, accuracy, disjointness-violation rate).
  3. [Section 4.2, Eq. (8)–(10)] The proposed loss is not fully well-defined as written. Eq. (8) indexes A_ref→tgt[u, v] with positions u and v, but the attention matrix is indexed by token IDs. While Eq. (9) defines a mapping G(·) from reference-point coordinates to global reference token indices, no analogous mapping is given for target-point coordinates v^(i,j) to target-latent token indices. Without this quantization/tokenization rule, Eq. (10) cannot be evaluated in implementation. Please specify how each v is converted to a target token index and how boundary cases are handled.
  4. [Section 5.2.1–5.3, Tables 1–3] All quantitative comparisons are reported as single-run point estimates without error bars or significance tests. Several claimed improvements are small (e.g., CLIP-I 84.30 vs. 83.50 in Table 1; DPG 96.55 vs. 96.93 in Table 2; CLIP-T 32.40 vs. 32.23 in Table 1). In view of the SOTA claims, please report at least three seeds with standard deviations for the main tables and ablations, and state whether the differences are statistically reliable. This is particularly important because Table 3 has only three conditions with no variance information.
minor comments (5)
  1. [Section 3, Eq. (1); Section 4.3, Eq. (11)] Notation is inconsistent: Eq. (1) defines C^(i,k) = {(u_i,j, v_i,j)}, but Eq. (11) refers to (u_j, v_j) ∈ C^(i,k). Please align the subscript conventions.
  2. [Section 4.3, Eq. (12)] The symbol '||·||' is used for normalization, but the exact norm (e.g., L2) and whether it is applied per-reference or per-token is not defined. Please clarify.
  3. [Abstract and Section 3] The dataset is described as 'meticulously annotated', but the pipeline is fully automatic (GPT-4o prompt generation, T2I synthesis, Lang-SAM segmentation, FLUX Kontext viewpoint correction). Please soften the wording or provide evidence of human verification of the correspondence labels.
  4. [References] References [20] and [21] are duplicates (both DreamO), and [35] and [36] are duplicates (both SSR-Encoder). Please merge or remove the redundant entries.
  5. [Fig. 6 and Section 5.3.2] The caption and text refer to the disentanglement loss as 'L_rd', while the main method section consistently uses 'L_MD'. Please unify the symbol.

Circularity Check

0 steps flagged

No significant circularity: the method is evaluated on external benchmarks and the proposed losses are not inversions of the evaluation metrics.

full rationale

The paper's central derivation is the training objective in Eq. 10 (semantic correspondence attention loss) and Eq. 14 (multi-reference disentanglement loss), combined in Eq. 15. These losses supervise attention maps using semantic correspondences from the SemAlign-MS training set. There is no place where a predicted evaluation quantity is defined in terms of a fitted parameter or where the benchmark metric appears in the training loss. Evaluations on DreamBench and XVerseBench use external metrics (CLIP-I, DINO, CLIP-T, DPG, ID-Sim, IP-Sim, AES) that are not optimized during training and are not used to set the hyperparameters α and β. The claim of 4+ subject fidelity is an empirical claim supported by qualitative results, not a definitional consequence of the loss equations; the absence of per-subject-count quantitative tables is an evidence gap, not circularity. The paper's only self-citation to prior work with overlapping authorship (MS-Diffusion [29]) appears in related work as a baseline and is not load-bearing for any derivation. No uniqueness theorem or ansatz is imported from the authors' own prior work. The automatically generated SemAlign-MS dataset may raise distributional generalization concerns, but that is a data-quality issue, not a circular reduction of the method's claims to its inputs.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The central claim rests on the correctness of the automatically built correspondence dataset, on the ability to steer DiT attention via the two losses, and on the transfer of that steering from training (where the target image is known) to generation (where it is not). The dataset and its construction are not released or fully specified, which is why these assumptions are the load-bearing ones. Two loss weights and the LoRA rank are hand-chosen without sensitivity analysis, adding three free parameters.

free parameters (3)
  • alpha (LSCA weight) = 0.4
    In Eq. 15, the semantic correspondence attention loss weight is set to 0.4 by hand; no sensitivity analysis is given.
  • beta (LMD weight) = 0.6
    In Eq. 15, the multi-reference disentanglement loss weight is set to 0.6 by hand; no sensitivity analysis is given.
  • LoRA rank = 128
    LoRA rank for the reference branch (Section 5.1.1); chosen without ablation.
axioms (5)
  • domain assumption The semantic correspondences in SemAlign-MS are correct and dense enough for attention supervision.
    Section 3 asserts correspondences are established and 'validated' but gives no algorithm, accuracy numbers, or human verification.
  • domain assumption Attention distributions in the DiT are meaningful and can be steered by cross-entropy supervision on known target positions.
    The SCAL loss (Eq. 10) operates on averaged attention across all blocks, assuming that raising attention at one token transfers to better image generation. No analysis of attention interpretability is provided.
  • domain assumption Training-time supervision using the known target latent transfers to inference when the target latent is generated from scratch.
    The losses are applied only while the real target is available; at inference there is no target to point to. The paper does not address this generalization gap.
  • domain assumption The base model (FLUX-1.0-DEV) can be finetuned with these auxiliary losses without semantic drift.
    Standard assumption for any finetune; not examined in the paper.
  • ad hoc to paper The disjointness constraint V^(i,k1) ∩ V^(i,k2) = ∅ (Eq. 2) is satisfiable for all training samples.
    If two references must map to the same target region (e.g., overlapping subjects), the dataset would either drop such samples or provide ambiguous supervision; the paper does not state how overlaps are handled.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MOSAIC: Multi-Subject Personalized Generation via Correspondence-Aware Alignment and Disentanglement." pith.science (2026). https://pith.science/paper/2O7IYVZM

@misc{pith2026250901977,
  author       = {Pith},
  title        = {Pith review of: MOSAIC: Multi-Subject Personalized Generation via Correspondence-Aware Alignment and Disentanglement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2O7IYVZM}},
  note         = {Machine review of arXiv:2509.01977}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Multi-subject personalized generation presents unique challenges in maintaining identity fidelity and semantic coherence when synthesizing images conditioned on multiple reference subjects. Existing methods often suffer from identity blending and attribute leakage due to inadequate modeling of how different subjects should interact within shared representation spaces. We present MOSAIC, a representation-centric framework that rethinks multi-subject generation through explicit semantic correspondence and orthogonal feature disentanglement. Our key insight is that multi-subject generation requires precise semantic alignment at the representation level - knowing exactly which regions in the generated image should attend to which parts of each reference. To enable this, we introduce SemAlign-MS, a meticulously annotated dataset providing fine-grained semantic correspondences between multiple reference subjects and target images, previously unavailable in this domain. Building on this foundation, we propose the semantic correspondence attention loss to enforce precise point-to-point semantic alignment, ensuring high consistency from each reference to its designated regions. Furthermore, we develop the multi-reference disentanglement loss to push different subjects into orthogonal attention subspaces, preventing feature interference while preserving individual identity characteristics. Extensive experiments demonstrate that MOSAIC achieves state-of-the-art performance on multiple benchmarks. Notably, while existing methods typically degrade beyond 3 subjects, MOSAIC maintains high fidelity with 4+ reference subjects, opening new possibilities for complex multi-subject synthesis applications.

discussion (0)

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

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Scaling Multi-Reference Image Generation with Dynamic Reward Optimization

    cs.CV 2026-06 unverdicted novelty 6.0

    Introduces OmniRef-Bench benchmark and DyRef two-stage framework using Difficulty-aware Advantage Reweighting and Discriminative Reward Scaling to improve open-source models on complex multi-reference image generation.

  2. DreamShot: Personalized Storyboard Synthesis with Video Diffusion Prior

    cs.CV 2026-04 unverdicted novelty 6.0

    DreamShot uses video diffusion priors and a role-attention consistency loss to produce coherent, personalized storyboards with better character and scene continuity than text-to-image methods.

  3. Vanast: Virtual Try-On with Human Image Animation via Synthetic Triplet Supervision

    cs.CV 2026-04 unverdicted novelty 6.0

    Vanast produces coherent garment-transferred human animation videos from a single human image, garment images, and pose guidance video using synthetic triplet supervision and a Dual Module video diffusion transformer ...

  4. Training-Free Image Editing with Visual Context Integration and Concept Alignment

    cs.CV 2026-04 unverdicted novelty 6.0

    VicoEdit performs training-free image editing by transforming source images directly with visual context and concept-alignment-guided posterior sampling, outperforming training-based methods.

  5. Hierarchical Concept-to-Appearance Guidance for Multi-Subject Image Generation

    cs.CV 2026-02 conditional novelty 6.0

    A diffusion-transformer framework with VLM-grounded masked attention and VAE dropout improves identity and prompt fidelity for multi-subject image generation.

  6. DreamShot: Personalized Storyboard Synthesis with Video Diffusion Prior

    cs.CV 2026-04 unverdicted novelty 5.0

    DreamShot generates controllable multi-shot storyboards from text or character references by exploiting video diffusion priors and a Role-Attention Consistency Loss.

Reference graph

Works this paper leans on

36 extracted references · 19 canonical work pages · cited by 5 Pith papers · 4 internal anchors

  1. [1]

    Surf: Speeded up robust features

    Herbert Bay, Tinne Tuytelaars, and Luc Van Gool. Surf: Speeded up robust features. InECCV, pages 404–417, 2006

    work page 2006

  2. [2]

    Xverse: Consistent multi-subject control of identity and semantic attributes via dit modulation.arXiv preprint arXiv:2506.21416, 2025

    Bowen Chen, Mengyi Zhao, Haomiao Sun, Li Chen, Xu Wang, Kang Du, and Xinglong Wu. Xverse: Consistent multi-subject control of identity and semantic attributes via dit modulation.arXiv preprint arXiv:2506.21416, 2025

    Pith/arXiv arXiv 2025

  3. [3]

    Arcface: Additive angular margin loss for deep face recognition

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. InCVPR, pages 4690–4699, 2019

    work page 2019

  4. [4]

    Aesthetic predictor v2.5: Siglip-based aesthetic score predictor.https://github.com/discus0434/ aesthetic-predictor-v2-5, 2024

    discus0434. Aesthetic predictor v2.5: Siglip-based aesthetic score predictor.https://github.com/discus0434/ aesthetic-predictor-v2-5, 2024. Accessed: 2024-12-08

    work page 2024

  5. [5]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In ICML, 2024

    2024

  6. [6]

    Proposal flow: Semantic correspondences from object proposals

    Bumsub Ham, Minsu Cho, Cordelia Schmid, and Jean Ponce. Proposal flow: Semantic correspondences from object proposals. IEEE TPMAI, 40:1711–1725, 2017

    work page 2017

  7. [7]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models. InICLR, 2022

    2022

  8. [8]

    Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135, 2024

    Xiwei Hu, Rui Wang, Yixiao Fang, Bin Fu, Pei Cheng, and Gang Yu. Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135, 2024

    Pith/arXiv arXiv 2024

  9. [9]

    Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, et al. Gpt-4o system card.arXiv preprint arXiv:2410.21276, 2024

    Pith/arXiv arXiv 2024

  10. [10]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. InICLR, 2015

    2015

  11. [11]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. InICCV, pages 4015–4026, 2023

    2023

  12. [12]

    Flux-dev-1.0.https://github.com/black-forest-labs/flux, 2024

    Black Forest Labs. Flux-dev-1.0.https://github.com/black-forest-labs/flux, 2024

    work page 2024

  13. [13]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742, 2025

    Pith/arXiv arXiv 2025

  14. [14]

    SFNet: Learning Object-aware Semantic Correspondence

    Junghyup Lee, Dohyung Kim, Jean Ponce, and Bumsub Ham. Sfnet: Learning object-aware semantic correspon- dence, 2019. URL https://arxiv.org/abs/1904.01810

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  15. [15]

    Reference-Based Sketch Image Colorization using Augmented-Self Reference and Dense Semantic Correspondence

    Junsoo Lee, Eungyeup Kim, Yunsung Lee, Dongjun Kim, Jaehyuk Chang, and Jaegul Choo. Reference-based sketch image colorization using augmented-self reference and dense semantic correspondence, 2020. URLhttps: //arxiv.org/abs/2005.05207

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  16. [16]

    Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing

    Dongxu Li, Junnan Li, and Steven Hoi. Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing. InNeurIPS, pages 30146–30166, 2023

    work page 2023

  17. [17]

    Distinctive image features from scale-invariant keypoints.IJCV, 60:91–110, 2004

    David G Lowe. Distinctive image features from scale-invariant keypoints.IJCV, 60:91–110, 2004

    work page 2004

  18. [18]

    RegionDrag: Fast Region-Based Image Editing with Diffusion Models

    Jingyi Lu, Xinghui Li, and Kai Han. Regiondrag: Fast region-based image editing with diffusion models, 2024. URL https://arxiv.org/abs/2407.18247

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  19. [19]

    Hyperpixel flow: Semantic correspondence with multi- layer neural features

    Juhong Min, Jongmin Lee, Jean Ponce, and Minsu Cho. Hyperpixel flow: Semantic correspondence with multi- layer neural features. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3395–3404, 2019

    work page 2019

  20. [21]

    Dreamo: A unified framework for image customization.arXiv preprint arXiv:2504.16915, 2025

    Chong Mou, Yanze Wu, Wenxu Wu, Zinan Guo, Pengze Zhang, Yufeng Cheng, Yiming Luo, Fei Ding, Shiwen Zhang, Xinghui Li, et al. Dreamo: A unified framework for image customization.arXiv preprint arXiv:2504.16915, 2025

    arXiv 2025

  21. [22]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023

    Pith/arXiv arXiv 2023

  22. [23]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In ICML, pages 8748–8763, 2021

    2021

  23. [24]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer, 2023. URL https://arxiv.org/abs/1910.10683

    Pith/arXiv arXiv 2023

  24. [25]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InCVPR, pages 22500–22510, 2023

    work page 2023

  25. [26]

    Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

    Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

    2024

  26. [27]

    Ominicontrol: Minimal and universal control for diffusion transformer

    Zhenxiong Tan, Songhua Liu, Xingyi Yang, Qiaochu Xue, and Xinchao Wang. Ominicontrol: Minimal and universal control for diffusion transformer. InICCV, 2025

    work page 2025

  27. [28]

    Emergent correspondence from image diffusion, 2023

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion, 2023. URLhttps://arxiv.org/abs/2306.03881

    Pith/arXiv arXiv 2023

  28. [29]

    Ms-diffusion: Multi-subject zero-shot image personalization with layout guidance

    Xierui Wang, Siming Fu, Qihan Huang, Wanggui He, and Hao Jiang. Ms-diffusion: Multi-subject zero-shot image personalization with layout guidance. InICLR, 2025

    work page 2025

  29. [30]

    Omnigen2: Exploration to advanced multimodal generation.arXiv preprint arXiv:2506.18871, 2025

    Chenyuan Wu, Pengfei Zheng, Ruiran Yan, Shitao Xiao, Xin Luo, Yueze Wang, Wanli Li, Xiyan Jiang, Yexin Liu, Junjie Zhou, et al. Omnigen2: Exploration to advanced multimodal generation.arXiv preprint arXiv:2506.18871, 2025

    Pith/arXiv arXiv 2025

  30. [31]

    Less-to-more generalization: Unlocking more controllability by in-context generation

    Shaojin Wu, Mengqi Huang, Wenxu Wu, Yufeng Cheng, Fei Ding, and Qian He. Less-to-more generalization: Unlocking more controllability by in-context generation. InICCV, 2025

    work page 2025

  31. [32]

    Ap-10k: A benchmark for animal pose estimation in the wild.arXiv preprint arXiv:2108.12617, 2021

    Hang Yu, Yufei Xu, Jing Zhang, Wei Zhao, Ziyu Guan, and Dacheng Tao. Ap-10k: A benchmark for animal pose estimation in the wild.arXiv preprint arXiv:2108.12617, 2021

    Pith/arXiv arXiv 2021

  32. [33]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence,

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Polania Cabrera, Varun Jampani, Deqing Sun, and Ming- Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence,

    work page

  33. [34]

    Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Eric Chen, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. Telling left from right: Identifying geometry-aware semantic correspondence, 2024. URLhttps://arxiv.org/ abs/2311.17034

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  34. [35]

    Ssr-encoder: Encoding selective subject representation for subject-driven generation

    Yuxuan Zhang, Yiren Song, Jiaming Liu, Rui Wang, Jinpeng Yu, Hao Tang, Huaxia Li, Xu Tang, Yao Hu, Han Pan, et al. Ssr-encoder: Encoding selective subject representation for subject-driven generation. InCVPR, pages 8069–8078, 2024

    work page 2024

  35. [36]

    Ssr-encoder: Encoding selective subject representation for subject-driven generation

    Yuxuan Zhang, Yiren Song, Jiaming Liu, Rui Wang, Jinpeng Yu, Hao Tang, Huaxia Li, Xu Tang, Yao Hu, Han Pan, et al. Ssr-encoder: Encoding selective subject representation for subject-driven generation. InCVPR, pages 8069–8078, 2024. 11

    work page 2024

  36. [2023]

    URL https://arxiv.org/abs/2305.15347

    Pith/arXiv arXiv

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.