FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization

Changshuo Wang; Lei Meng; Lei Wu; Mingzhe Yu; Xue Li; Yunshan Ma

REVIEW 2 major objections 5 minor 60 references

FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization

T0 review · 2 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read Fine-tuning a fashion outfit generator with direct preference optimization on multi-expert feedback produces more diverse, compatible, and personalized fashion items than supervised training.

desk verdict A plausible DPO-for-diffusion fine-tuning recipe with a multi-expert AI feedback module, but the personalization claim rests on a CLIP score that is both the training label and the evaluation metric. read the letter →

arxiv 2504.12900 v1 pith:N5I77JDN submitted 2025-04-17 cs.MM cs.IR

Mingzhe Yu , Yunshan Ma , Lei Wu , Changshuo Wang , Xue Li , Lei Meng This is my paper · ORCID

classification cs.MMcs.IR

keywords FashionoutfitgenerationimageGenerativerecommendationDirectpreferenceoptimizationMulti-expertfeedbackDiffusionmodelsPersonalizedfill-in-the-blank

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

Generative outfit models are normally trained to imitate a single ground-truth fashion item, which makes them shy away from equally valid alternatives. This paper argues that the better training signal is which of several generated candidates automated experts prefer, not which one matches a fixed label. FashionDPO fine-tunes the existing DiFashion generator with direct preference optimization, using preference pairs produced by three experts: a multimodal language model for image quality, a visual compatibility scorer, and a personalization score based on similarity between the generated item and the user's own fashion history. On iFashion and Polyvore-U, the fine-tuned model reports higher diversity, compatibility, and personalization than DiFashion and other baselines, and fashion-designer ratings favor it. If the claim holds, preference-based fine-tuning is a general recipe for making generative fashion recommenders diverge from ground truth without losing coherence.

What carries the argument

The load-bearing object is the multi-expert feedback pipeline paired with a DPO fine-tuning loop. Direct Preference Optimization is a training objective that upweights preferred outputs and downweights dispreferred ones relative to a frozen reference policy. For each generated candidate $i_{k,j,0}$, three scorers produce a weighted total: MiniCPM assigns a ten-level quality grade; a VBPR-style model scores visual compatibility between the candidate and the incomplete outfit; and a CLIP image encoder scores personalization as the cosine similarity between the candidate's embedding and the mean embedding of the user's same-category history. Candidates above the mean total score are labeled good, below are bad, and all within-outfit good-bad pairs enter the preference set. Fine-tuning then applies the DPO loss to the saved latents of each denoising timestep, with LoRA-restricted updates, so the pre-trained generator is pushed toward preferred candidates and away from dispreferred ones without training a reward model.

What would settle it

Run an independent preference study in which human raters or held-out click data compare FashionDPO and DiFashion outputs for the same users, and check whether human-preferred items score higher on Per than the model's choices; alternatively, retrain with a different personalization encoder and see whether the personalization gain of 60.39 versus 55.86 survives unchanged. If CLIP similarity to a mean embedding is not tracking true taste, the personalization improvement should shrink or reverse in such tests.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that direct preference optimization can replace supervised imitation in fashion outfit generation when the preference signal comes from several automatic experts rather than human labels. Concretely, starting from DiFashion, the framework generates several candidate items for an incomplete outfit, scores each with quality, compatibility, and personalization experts, forms good-bad pairs from the weighted scores, and fine-tunes the diffusion model with a DPO loss at every saved denoising timestep. The reported results on iFashion PFITB are Inception Score 33.80 versus 29.99, compatibility 0.74 versus 0.58, and personalization 60.39 versus 55.86 for DiFashion, with analogous gains on Polyvore-U and on the from-scratch GOR task. The paper reads these gains as evidence that the model learns broader styling principles and user alignment than ground-truth imitation allows.

Load-bearing premise

The load-bearing premise is that a similarity score between a generated garment image and the averaged image of the user's past same-category items captures that user's fashion taste; because the same score both labels training examples and measures personalization, the claimed personalization gain could be partly an artifact of training to the evaluator.

Editorial extensions

If this is right

If FashionDPO is right, a single fine-tuned model can serve both fill-in-the-blank and from-scratch outfit generation, since its reported scores are similar across the PFITB and GOR tasks.
The framework removes the need for task-specific reward models or new human annotations for each trend cycle, because feedback experts can be swapped or upgraded without redesigning the pipeline.
Because each expert contributes an independent signal, dropping any one of them measurably hurts its corresponding metric, so practitioners can diagnose which weakness needs a stronger expert.
Using preference pairs instead of ground-truth imitation increases output diversity, as reflected in the Inception Score jump from 29.99 to 33.80 on iFashion PFITB, which matters for users who want alternatives rather than one canonical item.

Reading between the lines

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

I infer that the reported personalization gain partly conflates training signal with evaluation metric: since the same CLIP similarity both labels preference pairs and defines Per, part of the 60.39 versus 55.86 gap may measure the model's fit to the metric rather than to human taste; an independent judgment test would separate these.
Beyond the paper, the same generate-candidates, score-with-multiple-experts, DPO-fine-tune loop should transfer to other creative generation tasks with one-of-many ground truth, such as interior staging, recipe pairing, or logo design, because it needs no task-specific reward model.
The ablation results suggest the experts are not equally easy to replace: swapping the compatibility scorer changes behavior more than swapping quality or personalization scorers, so investing in a stronger compatibility model is likely the highest-leverage next step.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: this is a legitimate application of DPO to fashion outfit generation—fine-tune DiFashion on preference pairs labeled by three off-the-shelf experts instead of ground-truth matching—and the diversity and compatibility results look plausible. The personalization pillar, though, is circular. Eq. (10) defines the personalization expert score as CLIP cosine similarity between the generated item and the averaged same-category history embedding. That same score labels the preference pairs in Eq. (13), and Section 5.1.3 reports the same cosine similarity as the Per. metric. The Per. gain (60.39 vs 55.86 on iFashion PFITB) is therefore partly a training-to-the-evaluator effect, not independent evidence about user taste. The mean-embedding aggregation in Eq. (6) makes this worse: any user with the same category history gets the same target, so the metric cannot distinguish genuine personalization from category-typical stylization.

Credit where it is earned. The multi-expert feedback module is a sensible recipe, and the quality and compatibility pillars are not circular: quality is trained with MiniCPM and evaluated with IS/IS-acc, compatibility is trained with VBPR and evaluated with OutfitGAN. The ablations in Table 2 and the expert-replacement study in Figure 6/Table 4 support the claim that each expert contributes. The human evaluation by fashion designers shows style, color, fabric, and variety gains over DiFashion, though it is not a per-user personalization test. Code and checkpoints are promised, which is good for reproducibility.

The soft spots beyond circularity are real but not fatal. Table 1 has no error bars or significance tests; the only significance numbers appear in Table 4 for the expert models, not for the headline results. Hyperparameters were selected by parameter search, and the paper does not say whether that search was on the test set—it looks like it was. The IS gain is suggestive but not a clean validation of the quality expert, since IS measures global diversity/clarity more than MiniCPM's stated fashion-design criteria. The human study is small: five designers, thirty outfit sets.

Overall, the diversity and compatibility claims are reasonably supported; the personalized-preference claim is not, until an independent personalization metric is used—something like a held-out interaction prediction task or a per-user human preference test with the actual users. This paper deserves a serious referee: the framework is useful, the application is new, and the circularity is fixable. I would accept it for peer review with the expectation of heavy revision on the personalization evaluation.

Referee Report

2 major / 5 minor

Summary. The paper proposes FashionDPO, a framework that fine-tunes a pre-trained fashion outfit generation model (DiFashion, itself based on Stable Diffusion) using direct preference optimization (DPO). The key idea is to replace supervised fine-tuning on ground-truth items with preference learning driven by automatically generated multi-expert feedback. Three experts are used: a quality expert based on MiniCPM, a compatibility expert based on VBPR, and a personalization expert based on CLIP cosine similarity between a generated item and the averaged same-category interaction history embedding. The experts' scores are combined into a weighted total, thresholded to label generated items as 'Good' or 'Bad', and the resulting preference pairs are used in a DPO loss to fine-tune the diffusion model. Experiments on iFashion and Polyvore-U compare FashionDPO against several baselines and report improvements in Inception Score (IS), IS-accuracy, compatibility (measured by the OutfitGAN discriminator), and personalization (measured by CLIP cosine similarity). The paper also includes ablations, a small expert human study, analyses of data/time cost, alternative expert implementations, and hyperparameter sensitivity.

Significance. If the reported gains are robust, FashionDPO offers a useful recipe for aligning generative fashion models with user preferences without hand-designed reward functions, and the code release supports reproducibility. The framework's generality is a strength: the three evaluation perspectives are modular and the authors show that experts can be replaced. The diversity (IS) and compatibility (Comp.) improvements rest on metrics that are not identical to the training feedback, which is good. However, the personalization pillar is weakened by a train-on-the-evaluator issue: the same CLIP score is used both to construct preference labels and to evaluate personalization, so the reported Per. gains are not independent evidence. In addition, the quantitative comparisons lack error bars and significance tests. The central claim therefore needs additional evidence to be fully supported.

major comments (2)

[Section 4.2, Eq. (10); Section 5.1.3; Table 1] The personalization expert score in Eq. (10) is the CLIP cosine similarity between a generated item and the averaged same-category history embedding, and the Per. evaluation metric in Section 5.1.3 is the same quantity. Because Eq. (11) incorporates this score into the weighted total that decides 'Good' vs 'Bad' labels, and the DPO loss in Eq. (18) is trained to increase the probability of the preferred item, the Per. improvements in Table 1 (e.g., 60.39 vs 55.86 on iFashion PFITB) can be at least partly attributed to training directly against the evaluation metric. This is not an independent measure of whether the generated items match an individual user's taste; it may simply reward stronger category-typical stylization. The paper should evaluate personalization with a metric that is not part of the training signal, for example a held-out user-item prediction model, a personalized retrieval task, or a human study that specifically tests whether generated items reflect the target user's style.
[Table 1 and Table 2] All quantitative results are point estimates without variance, confidence intervals, or significance tests. Diffusion sampling and LoRA fine-tuning are stochastic, and some reported gains are small (e.g., IS-acc 0.90 to 0.91, Polyvore-U Comp. 0.80 to 0.83). To substantiate the claim that FashionDPO consistently outperforms DiFashion and other baselines, the paper should report means and standard deviations over multiple seeds or bootstrap resampling, and where appropriate apply paired significance tests over the test outfits.

minor comments (5)

[Section 4.2, Eq. (11)] The min-max normalization norm(·) is not defined over which set of scores; please specify whether it is computed per-outfit over the m generated candidates or globally across the dataset.
[Section 5.1.4 and Eq. (18)] The implementation states that the loss LDPO is computed and gradients are updated at each timestep, but it is unclear whether the final loss is a sum, an average, or a randomly sampled timestep loss over T steps. Please clarify the exact training procedure.
[Section 5.3.3, Table 3] The human evaluation uses five fashion experts and 30 sets, but the paper does not describe how the 30 sets were sampled, whether each expert evaluated both models for the same incomplete outfits, or whether there was any inter-annotator agreement check. Reporting Fleiss' kappa or a similar measure would strengthen this evidence.
[Section 5.4.2, Table 4] The 'Accuracy' metric in Table 4 is not defined. The paper should state the ground truth used to compute accuracy for each expert (e.g., human labels, agreement with another expert, or an external gold standard) so the reader can interpret the numbers.
[Contributions list and Eq. (2)] There is a typo in the contributions list ('mulitple' should be 'multiple'). Also, Eq. (2) uses beta_1 and beta_2 without defining them, while Section 5.1.4 introduces beta_w and beta_l; please unify the notation.

Circularity Check

1 steps flagged · score 6.0 of 10

The personalization evaluation metric is the same CLIP cosine score used to label the DPO preference pairs, so the reported Per. gains are partly trained-to-the-evaluator.

fitted input called prediction [Section 4.2, Eq. (10)-(13); Section 5.1.3; Table 1]
"Personalization. ... s_p = CLIP_Score(v_k, h_k) = (v_k·h_k)/(||v_k||||h_k||). (10) ... Personalization (Per.): We use the foundation model CLIP [31] to extract the image embeddings of the items that a user has interacted in the history. Then we calculate the cosine similarity between generated fashion items and history image embeddings."

The personalization expert score in Eq. (10) is a CLIP cosine similarity between a generated item and the user's history condition. This score enters the weighted total in Eq. (11), which determines whether an item is labeled 'Good' or 'Bad' in Eq. (13), and the DPO loss in Eq. (18) is optimized to increase that score. Section 5.1.3 defines the Per. evaluation metric as the same CLIP cosine similarity between generated items and history image embeddings, and Table 1 reports Per. as evidence of personalization. Consequently, the reported Per.

full rationale

The central circular step is confined to the personalization pillar. Eq. (10) defines the personalization expert as a CLIP cosine similarity to the history condition, Eq. (13) turns that score into DPO preference labels, and Section 5.1.3 reports the same type of CLIP cosine similarity as the Per. metric. Thus the model is explicitly fine-tuned to increase the quantity later reported as a personalization evaluation, making that portion of the headline result partly circular by construction. The other two pillars are evaluated with different instruments than their training experts: quality uses MiniCPM feedback but Inception Score evaluation, and compatibility uses VBPR feedback but the OutfitGAN discriminator for evaluation, so those claims retain independent content. The human evaluation in Table 3 was performed by fashion experts rating style, color, fabric, and variety, which supports compatibility and diversity but does not independently validate user-specific personalization. Self-citations to DiFashion and related prior work are not load-bearing for a circular argument; DiFashion is used as a base model, and the DPO framework is grounded in standard external references. Overall, the circularity is real but partial, affecting one of the three claimed evaluation perspectives.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The framework rests on assumptions about the validity of three AI experts and the DPO per-timestep assumption; the personalization expert doubles as the evaluation metric, which is the main circularity. No new physical entities are introduced. Several hyperparameters are chosen by hand or by test-set performance.

free parameters (6)

beta (beta_w, beta_l) = 0.5
DPO deviation control; tuned by parameter search using evaluation metrics (Section 5.4.3), so it is fitted to test data.
expert weights alpha_q, alpha_c, alpha_p = 1, 1, 1
Weights in Eq. 11 for combining quality, compatibility, and personalization scores; chosen by hand with no sensitivity analysis reported.
preference threshold t = average of scores
Used to label Good/Bad items; set to the mean of scores for each outfit, a heuristic with no justification.
number of generated candidates m = 7
Number of items generated per outfit for preference pair construction; no sensitivity analysis reported.
training subset size = 1000 outfits
Selected because Figure 5 shows 1,000 gives similar performance to 2,000 at half the time; this is test-performance-based model selection.
number of epochs = 5
Fine-tuning runs for five epochs, chosen based on performance curves in Figures 5-7.

assumptions (6)

domain assumption If a final generated image is preferred, then the state and action at every intermediate denoising timestep are also preferred.
Adopted from D3PO; used in Section 4.3 to apply the DPO loss at every timestep. If early noisy states are not comparable, the per-timestep loss is misdirected.
domain assumption CLIP cosine similarity between a generated item and the average latent of the user's same-category history is a valid personalization signal.
Used in Eq. 10 both to create preference pairs and as the Per evaluation metric; no independent validation of this proxy.
domain assumption VBPR trained with BPR on outfit-item matching accurately measures fashion compatibility.
Used as the compatibility expert; biased scores would corrupt the preference labels.
domain assumption MiniCPM's 10-level quality scores reflect human-perceived fashion image quality.
Used as the quality expert; the paper's Table 4 tests accuracy against GPT-4 and human labels, but only partially.
domain assumption The Bradley-Terry preference model and the DPO derivation from Rafailov et al. apply to diffusion model outputs.
Borrowed from DPO literature; needed for the loss in Eq. 18.
domain assumption The pretrained DiFashion model provides a valid base policy and reference model.
FashionDPO fine-tunes and compares against DiFashion; any deficiency in the base model transfers.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization." pith.science (2026). https://pith.science/paper/N5I77JDN

@misc{pith2026250412900,
  author       = {Pith},
  title        = {Pith review of: FashionDPO:Fine-tune Fashion Outfit Generation Model using Direct Preference Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N5I77JDN}},
  note         = {Machine review of arXiv:2504.12900}
}

read the original abstract

Personalized outfit generation aims to construct a set of compatible and personalized fashion items as an outfit. Recently, generative AI models have received widespread attention, as they can generate fashion items for users to complete an incomplete outfit or create a complete outfit. However, they have limitations in terms of lacking diversity and relying on the supervised learning paradigm. Recognizing this gap, we propose a novel framework FashionDPO, which fine-tunes the fashion outfit generation model using direct preference optimization. This framework aims to provide a general fine-tuning approach to fashion generative models, refining a pre-trained fashion outfit generation model using automatically generated feedback, without the need to design a task-specific reward function. To make sure that the feedback is comprehensive and objective, we design a multi-expert feedback generation module which covers three evaluation perspectives, \ie quality, compatibility and personalization. Experiments on two established datasets, \ie iFashion and Polyvore-U, demonstrate the effectiveness of our framework in enhancing the model's ability to align with users' personalized preferences while adhering to fashion compatibility principles. Our code and model checkpoints are available at https://github.com/Yzcreator/FashionDPO.

Figures

Figures reproduced from arXiv: 2504.12900 by the authors.

**Figure 2.** The overview of FashionDPO, which consists of three consecutive key modules: 1) Fashion Image Generation without [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Epoch-wise comparison of FashionDPO’s performance across different fine-tuning epochs. As epochs increase, the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Model-wise comparison of different models’ gener [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: We fine-tune our model on a subset with 𝑛 outfits, where 𝑛 ∈ {100, 500, 800, 1000, 2000}, to explore the impact of varying datase size on model performance. Lines represent models fine-tuned on different subsets, with the x-axis as epochs and the y-axis as the evaluati…

**Figure 6.** Figure 6: The impact of expert implementations. Each line [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Effects of the hyper-parameter 𝛽 in controlling the deviation. Bars show the scores on the evaluation metric after fine-tuning for five epochs using different hyper-parameters. compared to MiniCPM, as evidenced by GPT-4 assigning relatively high scores to negative item…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 31 canonical work pages

[1]

Wen Chen, Pipei Huang, Jiaming Xu, Xin Guo, Cheng Guo, Fei Sun, Chao Li, Andreas Pfadler, Huan Zhao, and Binqiang Zhao. 2019. POG: Personalized Outfit Generation for Fashion Recommendation at Alibaba iFashion. In KDD. ACM, 2662–2670

2019
[2]

Yujuan Ding, Zhihui Lai, P. Y. Mok, and Tat-Seng Chua. 2024. Computational Technologies for Fashion Recommendation: A Survey. ACM Comput. Surv. 56, 5 (2024), 121:1–121:45

work page 2024
[3]

Yujuan Ding, Yunshan Ma, Wenqi Fan, Yige Yao, Tat-Seng Chua, and Qing Li
[4]

Yujuan Ding, Yunshan Ma, Lizi Liao, Wai Keung Wong, and Tat-Seng Chua. 2022. Leveraging Multiple Relations for Fashion Trend Forecasting Based on Social Media. IEEE Trans. Multim. 24 (2022), 2287–2299

work page 2022
[5]

Yujuan Ding, P. Y. Mok, Yunshan Ma, and Yi Bin. 2023. Personalized fashion outfit generation with user coordination preference learning. Inf. Process. Manag. 60, 5 (2023), 103434

work page 2023
[6]

Xue Dong, Xuemeng Song, Fuli Feng, Peiguang Jing, Xin-Shun Xu, and Liqiang Nie. 2019. Personalized Capsule Wardrobe Creation with Garment and User Modeling. In ACM Multimedia. ACM, 302–310

work page 2019
[7]

Xiaoyu Du, Kun Qian, Yunshan Ma, and Xinguang Xiang. 2023. Enhancing item-level bundle representation for bundle recommendation. ACM Transactions on Recommender Systems (2023)

work page 2023
[8]

Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang

Show all 60 references

[9]

Xintong Han, Zuxuan Wu, Yu-Gang Jiang, and Larry S. Davis. 2017. Learning Fashion Compatibility with Bidirectional LSTMs. In ACM Multimedia. ACM, 1078–1086

2017
[10]

Xintong Han, Zuxuan Wu, Zhe Wu, Ruichi Yu, and Larry S. Davis. 2018. VITON: An Image-Based Virtual Try-On Network. InCVPR. Computer Vision Foundation / IEEE Computer Society, 7543–7552

2018
[11]

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2015. Deep Residual Learning for Image Recognition. CoRR abs/1512.03385 (2015)

2015 arXiv
[12]

Ruining He and Julian J. McAuley. 2016. VBPR: Visual Bayesian Personalized Ranking from Implicit Feedback. In AAAI. AAAI Press, 144–150

2016
[13]

Jonathan Ho and Tim Salimans. 2022. Classifier-Free Diffusion Guidance. CoRR abs/2207.12598 (2022)

2022 arXiv
[14]

Qihan Huang, Long Chan, Jinlong Liu, Wanggui He, Hao Jiang, Mingli Song, and Jie Song. 2024. PatchDPO: Patch-level DPO for Finetuning-free Personalized Image Generation. CoRR abs/2412.03177 (2024)

2024 arXiv
[15]

Cong Phuoc Huynh, Arri Ciptadi, Ambrish Tyagi, and Amit Agrawal. 2018. CRAFT: Complementary Recommendations Using Adversarial Feature Trans- former. CoRR abs/1804.10871 (2018)

2018 arXiv
[16]

Jeongho Kim, Gyojung Gu, Minho Park, Sunghyun Park, and Jaegul Choo. 2023. StableVITON: Learning Semantic Correspondence with Latent Diffusion Model for Virtual Try-On. CoRR abs/2312.01725 (2023)

2023 arXiv
[17]

Hyeon-Ju Lee and Seok-Jun Buu. 2024. Deep Generative Replay With Denoising Diffusion Probabilistic Models for Continual Learning in Audio Classification. IEEE Access 12 (2024), 134714–134727

2024
[18]

Harrison Lee, Samrat Phatale, Hassan Mansoor, Kellie Lu, Thomas Mesnard, Colton Bishop, Victor Carbune, and Abhinav Rastogi. 2023. RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. CoRR abs/2309.00267 (2023)

2023 arXiv
[19]

Harrison Lee, Samrat Phatale, Hassan Mansoor, Thomas Mesnard, Johan Ferret, Kellie Lu, Colton Bishop, Ethan Hall, Victor Carbune, Abhinav Rastogi, and Sushant Prakash. 2024. RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. In ICML. OpenReview.net

2024
[20]

Xingchen Li, Xiang Wang, Xiangnan He, Long Chen, Jun Xiao, and Tat-Seng Chua. 2020. Hierarchical Fashion Graph Network for Personalized Outfit Recom- mendation. In SIGIR. ACM, 159–168

2020
[21]

Zhanhao Liang, Yuhui Yuan, Shuyang Gu, Bohan Chen, Tiankai Hang, Ji Li, and Liang Zheng. 2024. Step-aware Preference Optimization: Aligning Preference with Denoising Performance at Each Step. CoRR abs/2406.04314 (2024)

2024 arXiv
[22]

Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. 2022. Pseudo Numerical Methods for Diffusion Models on Manifolds. In ICLR. OpenReview.net

2022
[23]

Xiaohao Liu, Jie Wu, Zhulin Tao, Yunshan Ma, Yinwei Wei, and Tat-Seng Chua
[24]

Zhi Lu, Yang Hu, Yan Chen, and Bing Zeng. 2021. Personalized Outfit Recom- mendation With Learnable Anchors. In CVPR. Computer Vision Foundation / IEEE, 12722–12731

2021
[25]

Zhi Lu, Yang Hu, Yunchao Jiang, Yan Chen, and Bing Zeng. 2019. Learning Binary Code for Personalized Fashion Recommendation. In CVPR. Computer Vision Foundation / IEEE, 10562–10570

2019
[26]

Yunshan Ma, Yujuan Ding, Xun Yang, Lizi Liao, Wai Keung Wong, and Tat-Seng Chua. 2020. Knowledge Enhanced Neural Fashion Trend Forecasting. In ICMR. ACM, 82–90

2020
[27]

Yunshan Ma, Yingzhi He, Xiang Wang, Yinwei Wei, Xiaoyu Du, Yuyangzi Fu, and Tat-Seng Chua. 2024. MultiCBR: Multi-view Contrastive Learning for Bundle Recommendation. ACM Trans. Inf. Syst. 42, 4 (2024), 100:1–100:23

2024
[28]

Maryam Moosaei, Yusan Lin, Ablaikhan Akhazhanov, Huiyuan Chen, Fei Wang, and Hao Yang. 2022. OutfitGAN: Learning Compatible Items for Generative Fashion Outfits. In CVPR Workshops. IEEE, 2272–2276

2022
[29]

Sanghyeon Na, Yonggyu Kim, and Hyunjoon Lee. 2024. Boost Your Own Human Image Generation Model via Direct Preference Optimization with AI Feedback. CoRR abs/2405.20216 (2024)

2024 arXiv
[30]

Gaurav Parmar, Richard Zhang, and Jun-Yan Zhu. 2022. On Aliased Resizing and Surprising Subtleties in GAN Evaluation. In CVPR. IEEE, 11400–11410

2022
[31]

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2021. Learning Transferable Visual Models From Natural Language Supervision. In ICML (Proce...

2021
[32]

Manning, Stefano Ermon, and Chelsea Finn

Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct Preference Optimization: Your Language Model is Secretly a Reward Model. In NeurIPS

2023
[33]

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme
[34]

Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-Resolution Image Synthesis with Latent Diffusion Models. In CVPR. IEEE, 10674–10685

2022
[35]

Yong-Siang Shih, Kai-Yueh Chang, Hsuan-Tien Lin, and Min Sun. 2018. Com- patibility Family Learning for Item Recommendation and Generation. In AAAI. AAAI Press, 2403–2410

2018
[36]

David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Pan- neershelvam, Marc Lanctot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicra...

2016
[37]

Lillicrap, Fan Hui, Laurent Sifre, George van den Driess- che, Thore Graepel, and Demis Hassabis

David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, Yutian Chen, Timothy P. Lillicrap, Fan Hui, Laurent Sifre, George van den Driess- che, Thore Graepel, and Demis Hassabis. 201...

2017
[38]

Jiaming Song, Chenlin Meng, and Stefano Ermon. 2021. Denoising Diffusion Implicit Models. In ICLR. OpenReview.net

2021
[39]

Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbig- niew Wojna. 2016. Rethinking the Inception Architecture for Computer Vision. In CVPR. IEEE Computer Society, 2818–2826

2016
[40]

Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik
[41]

Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, and Wangmeng Zuo
[42]

Zhenyu Xie, Zaiyu Huang, Xin Dong, Fuwei Zhao, Haoye Dong, Xijin Zhang, Feida Zhu, and Xiaodan Liang. 2023. GP-VTON: Towards General Purpose Virtual Try-On via Collaborative Local-Flow Global-Parsing Learning. In CVPR. IEEE, 23550–23559

2023
[43]

Yiyan Xu, Wenjie Wang, Fuli Feng, Yunshan Ma, Jizhi Zhang, and Xiangnan He
[44]

Yuhui Xu, Lingxi Xie, Xiaotao Gu, Xin Chen, Heng Chang, Hengheng Zhang, Zhengsu Chen, Xiaopeng Zhang, and Qi Tian. 2024. QA-LoRA: Quantization- Aware Low-Rank Adaptation of Large Language Models. In ICLR. OpenRe- view.net

2024
[45]

CoRR abs/2311.12908 (2023)

Diffusion Model Alignment Using Direct Preference Optimization. CoRR abs/2311.12908 (2023)

2023 arXiv
[46]

Zilin Yang, Zhuo Su, Yang Yang, and Ge Lin. 2018. From recommendation to generation: A novel fashion clothing advising framework. In2018 7th International Conference on Digital Home (ICDH) . IEEE, 180–186

2018
[47]

ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation. In ICCV. IEEE, 15897–15907
[48]

Mingzhe Yu, Yunshan Ma, Lei Wu, Kai Cheng, Xue Li, Lei Meng, and Tat-Seng Chua. 2024. Smart Fitting Room: A One-stop Framework for Matching-aware Virtual Try-On. In ICMR. ACM, 184–192

2024
[49]

Patel, Haochen Wang, Xun Huang, Ting-Chun Wang, Ming-Yu Liu, and Yogesh Balaji

Yu Zeng, Vishal M. Patel, Haochen Wang, Xun Huang, Ting-Chun Wang, Ming-Yu Liu, and Yogesh Balaji. 2024. JeDi: Joint-Image Diffusion Models for Finetuning- Free Personalized Text-to-Image Generation. In CVPR. IEEE, 6786–6795. FashionDPO: Fine-tune Fashion Outfit Generation Mod...

2024
[50]

In SIGIR

Diffusion Models for Generative Outfit Recommendation. In SIGIR. ACM, 1350–1359
[51]

Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. 2023. Adding Conditional Control to Text-to-Image Diffusion Models. In ICCV. IEEE, 3813–3824

2023
[52]

Kai Yang, Jian Tao, Jiafei Lyu, Chunjiang Ge, Jiaxin Chen, Qimai Li, Weihan Shen, Xiaolong Zhu, and Xiu Li. 2023. Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model. CoRR abs/2311.13231 (2023)

2023 arXiv
[54]

Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, et al. 2024. MiniCPM-V: A GPT-4V Level MLLM on Your Phone. arXiv preprint arXiv:2408.01800 (2024)

2024 arXiv
[57]

Huijing Zhan, Jie Lin, Kenan Emir Ak, Boxin Shi, Ling-Yu Duan, and Alex C. Kot. 2022. $Aˆ3$-FKG: Attentive Attribute-Aware Fashion Knowledge Graph for Outfit Preference Prediction. IEEE Trans. Multim. 24 (2022), 819–831

2022
[59]

Efros, Eli Shechtman, and Oliver Wang

Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang
[2009]

BPR: Bayesian Personalized Ranking from Implicit Feedback. In UAI. AUAI Press, 452–461
[2018]

The Unreasonable Effectiveness of Deep Features as a Perceptual Metric. In CVPR. Computer Vision Foundation / IEEE Computer Society, 586–595
[2023]

In ACM Multimedia

Taming the Power of Diffusion Models for High-Quality Virtual Try-On with Appearance Flow. In ACM Multimedia. ACM, 7599–7607
[2024]

In WWW (Companion Volume)

FashionReGen: LLM-Empowered Fashion Report Generation. In WWW (Companion Volume). ACM, 991–994
[2025]

In KDD (1)

Fine-tuning Multimodal Large Language Models for Product Bundling. In KDD (1). ACM, 848–858

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [1]

Wen Chen, Pipei Huang, Jiaming Xu, Xin Guo, Cheng Guo, Fei Sun, Chao Li, Andreas Pfadler, Huan Zhao, and Binqiang Zhao. 2019. POG: Personalized Outfit Generation for Fashion Recommendation at Alibaba iFashion. In KDD. ACM, 2662–2670

2019

[2] [2]

Yujuan Ding, Zhihui Lai, P. Y. Mok, and Tat-Seng Chua. 2024. Computational Technologies for Fashion Recommendation: A Survey. ACM Comput. Surv. 56, 5 (2024), 121:1–121:45

work page 2024

[3] [3]

Yujuan Ding, Yunshan Ma, Wenqi Fan, Yige Yao, Tat-Seng Chua, and Qing Li

[4] [4]

Yujuan Ding, Yunshan Ma, Lizi Liao, Wai Keung Wong, and Tat-Seng Chua. 2022. Leveraging Multiple Relations for Fashion Trend Forecasting Based on Social Media. IEEE Trans. Multim. 24 (2022), 2287–2299

work page 2022

[5] [5]

Yujuan Ding, P. Y. Mok, Yunshan Ma, and Yi Bin. 2023. Personalized fashion outfit generation with user coordination preference learning. Inf. Process. Manag. 60, 5 (2023), 103434

work page 2023

[6] [6]

Xue Dong, Xuemeng Song, Fuli Feng, Peiguang Jing, Xin-Shun Xu, and Liqiang Nie. 2019. Personalized Capsule Wardrobe Creation with Garment and User Modeling. In ACM Multimedia. ACM, 302–310

work page 2019

[7] [7]

Xiaoyu Du, Kun Qian, Yunshan Ma, and Xinguang Xiang. 2023. Enhancing item-level bundle representation for bundle recommendation. ACM Transactions on Recommender Systems (2023)

work page 2023

[8] [8]

Junhong Gou, Siyu Sun, Jianfu Zhang, Jianlou Si, Chen Qian, and Liqing Zhang