arxiv: 2604.16892 · v1 · submitted 2026-04-18 · 💻 cs.CV

Recognition: unknown

CrossFlowDG: Bridging the Modality Gap with Cross-modal Flow Matching for Domain Generalization

Antonios Kritikos , Nikolaos Spanos , Athanasios Voulodimos

Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:42 UTC · model grok-4.3

classification 💻 cs.CV

keywords domain generalizationflow matchingcross-modal alignmentmodality gapCLIP embeddingsimage-to-text transportTerraIncognitaVMamba

0 comments

The pith

CrossFlowDG transports domain-biased image embeddings to class-matching text embeddings via cross-modal flow matching to close the modality gap in domain generalization.

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

Domain generalization fails when vision models latch onto stylistic variations instead of stable class semantics. Multimodal methods try to anchor images to text embeddings for invariance, yet cosine-similarity alignment still leaves image and text vectors geometrically separated. CrossFlowDG trains a continuous flow that moves each image embedding directly toward the correct text embedding in the shared space, using no added noise. The resulting features yield competitive accuracy across standard DG benchmarks and state-of-the-art results on TerraIncognita when paired with VMamba image encoding and CLIP text encoding.

Core claim

The framework learns a continuous transformation in the joint Euclidean latent space that explicitly transports domain-biased image embeddings toward domain-invariant text embeddings of the correct class, using noise-free cross-modal flow matching to close the residual modality gap left by cosine similarity-based contrastive alignment.

What carries the argument

Cross-modal flow matching that learns a continuous transport map from image embeddings to matching text embeddings in the shared latent space.

Load-bearing premise

CLIP text embeddings serve as truly domain-invariant anchors and the learned flow preserves semantic content without introducing new biases or artifacts.

What would settle it

If transported image embeddings no longer retrieve the correct class label when matched to text embeddings on a held-out domain shift, the claim that the flow maintains semantics would be falsified.

Figures

Figures reproduced from arXiv: 2604.16892 by Antonios Kritikos, Athanasios Voulodimos, Nikolaos Spanos.

**Figure 1.** Figure 1: Overview of the CrossFlowDG framework. Image samples are randomly paired with stylistic descriptions from the Textual Domain Bank (TDB). Then, the class corresponding to the image is appended to the textual description, and the unimodal inputs are fed to pretrained encoders. These intermediate representations are projected through VAEs in a joint latent space, which is formed via interand intra-modal cosi… view at source ↗

**Figure 2.** Figure 2: Plots of our proposed metrics (dAMG, dRMG, dCA in each row) and t-SNE visualization of the latent space of the L100 target domain of the TerraIncognita dataset across five experiments. Notes: In the legend, “img” denotes the initial image latent (as encoded by the VAE), while “flow” denotes the resulting latent after flowing (or simple ResNet mapping in the case of (d)). For the dRMG metric, the plot is ad… view at source ↗

read the original abstract

Domain generalization (DG) aims to maintain performance under domain shift, which in computer vision appears primarily as stylistic variations that cause models to overfit to domain-specific appearance cues rather than class semantics. To overcome this, recent methods use textual representations as stable, domain-invariant anchors. However, multimodal approaches that rely on cosine similarity-based contrastive alignment leave a modality gap where image and text embeddings remain geometrically separated despite semantic correspondence. We propose CrossFlowDG, a novel DG framework that addresses this residual gap using noise-free, cross-modal flow matching. By learning a continuous transformation in the joint Euclidean latent space, our framework explicitly transports domain-biased image embeddings toward domain-invariant text embeddings of the correct class. Using the efficient VMamba image encoder and CLIP's text encoder, CrossFlowDG is tested against four common DG benchmarks, and achieves competitive performance on several benchmarks and state-of-the-art on TerraIncognita. Code is available at: https://github.com/ajkrit/CrossFlowDG

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.

Desk Editor's Note private letter to a colleague

CrossFlowDG uses deterministic noise-free flow matching to transport image embeddings onto fixed CLIP text embeddings as domain-invariant targets, with code released and claimed SOTA on one benchmark.

read the letter

The main point is that this paper trains a flow to move domain-biased image embeddings straight toward class-matched CLIP text embeddings in a shared space, after initial contrastive alignment. The flow is noise-free and deterministic, following straight paths, and runs end-to-end with a VMamba image encoder on source domains only. The downstream classifier then works on the transported features for unseen target domains. This is a direct attempt to close the geometric modality gap that cosine-based contrastive methods leave behind. The code is public, which helps with checking the implementation details. The setup earns credit for sticking to standard DG benchmarks and protocols without obvious circularity in the argument. The text embeddings serve as fixed anchors, and the flow operates in Euclidean space, so the high-level construction is internally consistent. Results are reported as competitive across four benchmarks and state-of-the-art on TerraIncognita. The soft spots are limited. The approach assumes CLIP text embeddings stay sufficiently domain-invariant and that the flow does not distort class semantics or add new artifacts during transport. If either assumption slips, the gains could shrink. The abstract does not include ablations that isolate the flow's contribution from the backbone or base alignment, so the full paper's tables and error analysis matter for confirming the transport is the active ingredient. No load-bearing inconsistencies appear in the described method. This is for readers working on multimodal domain generalization or flow-based alignment in vision-language models. Someone tracking practical ways to reduce modality gaps beyond contrastive losses would find the specific formulation and benchmark numbers useful. I would send it to peer review. The core idea is coherent, the code adds reproducibility value, and referees can verify the experiments and any edge cases around the flow's generalization.

Referee Report

2 major / 2 minor

Summary. The paper proposes CrossFlowDG, a domain generalization (DG) method that learns a noise-free cross-modal flow to transport domain-biased image embeddings (from VMamba) toward fixed, class-matched CLIP text embeddings in a shared Euclidean space. Trained end-to-end on source domains only, the framework aims to close the residual modality gap left by contrastive alignment while preserving semantics, and reports competitive results on standard DG benchmarks with state-of-the-art performance on TerraIncognita.

Significance. If the empirical claims hold, the work is significant for introducing deterministic flow matching as an explicit transport mechanism in multimodal DG, moving beyond cosine-similarity alignment. The noise-free, straight-path design and use of an efficient VMamba encoder are strengths; the released code supports reproducibility and allows independent verification of the transport's effect on domain invariance.

major comments (2)

[§5] §5 (Experiments): The abstract asserts competitive performance across four DG benchmarks and SOTA on TerraIncognita, yet the manuscript must include full quantitative tables (with per-domain accuracies, means, and standard deviations over multiple seeds) and direct numerical comparisons to the strongest baselines; without these, the magnitude and reliability of the claimed improvement cannot be assessed.
[§3] §3 (Method): The central claim that the learned flow preserves class semantics during transport rests on the unverified assumption that CLIP text embeddings serve as unbiased anchors; the paper should add an ablation or diagnostic (e.g., nearest-neighbor class consistency or t-SNE before/after transport) to show that the ODE integration does not introduce new inter-class confusion or domain-specific artifacts.

minor comments (2)

[§3] Notation in the flow-matching formulation should explicitly define the velocity field and integration time interval to facilitate exact reproduction from the released code.
[§2] The related-work section would benefit from a concise comparison table contrasting CrossFlowDG with prior multimodal DG methods that also use CLIP text anchors.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for major revision. We address each major comment below, indicating the revisions we will make to strengthen the manuscript.

read point-by-point responses

Referee: [§5] §5 (Experiments): The abstract asserts competitive performance across four DG benchmarks and SOTA on TerraIncognita, yet the manuscript must include full quantitative tables (with per-domain accuracies, means, and standard deviations over multiple seeds) and direct numerical comparisons to the strongest baselines; without these, the magnitude and reliability of the claimed improvement cannot be assessed.

Authors: We agree that comprehensive quantitative reporting is necessary to substantiate the performance claims. The manuscript currently presents mean accuracies across the four DG benchmarks with comparisons to baselines. In the revised version, we will expand the tables to report per-domain accuracies, include standard deviations computed over multiple random seeds, and provide explicit numerical comparisons to the strongest baselines for each benchmark. revision: yes
Referee: [§3] §3 (Method): The central claim that the learned flow preserves class semantics during transport rests on the unverified assumption that CLIP text embeddings serve as unbiased anchors; the paper should add an ablation or diagnostic (e.g., nearest-neighbor class consistency or t-SNE before/after transport) to show that the ODE integration does not introduce new inter-class confusion or domain-specific artifacts.

Authors: We acknowledge that empirical verification of semantic preservation during transport strengthens the central claim. While the framework relies on class-matched CLIP text embeddings as anchors, we will add the suggested diagnostics in the revision: nearest-neighbor class consistency metrics and t-SNE visualizations of image embeddings before and after ODE integration, to demonstrate that the process does not introduce inter-class confusion or domain-specific artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents CrossFlowDG as an independent framework that learns a deterministic cross-modal flow to transport image embeddings toward fixed CLIP text embeddings using standard flow matching in a shared Euclidean space. No equations, derivations, or self-referential constructions are described that reduce the claimed transport or performance gains to fitted parameters by construction, self-citations as load-bearing premises, or ansatzes smuggled from prior author work. The method is trained end-to-end on source domains only with VMamba and CLIP encoders, evaluated on external DG benchmarks, and supported by released code. The derivation chain remains self-contained without reducing to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are explicitly introduced or quantified in the abstract; the method relies on standard flow-matching machinery and pre-trained encoders.

pith-pipeline@v0.9.0 · 5482 in / 1057 out tokens · 40789 ms · 2026-05-10T07:42:11.573733+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

58 extracted references · 7 canonical work pages · 2 internal anchors

[1]

Domain-adversarial neural networks

Hana Ajakan, Pascal Germain, Hugo Larochelle, Franc ¸ois Laviolette, and Mario Marchand. Domain-adversarial neural networks.arXiv preprint arXiv:1412.4446, 2014. 2

work page arXiv 2014
[2]

Albergo and Eric Vanden-Eijnden

Michael S. Albergo and Eric Vanden-Eijnden. Building nor- malizing flows with stochastic interpolants, 2023. 2

2023
[3]

Albergo, Nicholas M

Michael S. Albergo, Nicholas M. Boffi, and Eric Vanden- Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions, 2023. 2

2023
[4]

Recognition in terra incognita

Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. InProceedings of the European confer- ence on computer vision (ECCV), pages 456–473, 2018. 4

2018
[5]

Dgfamba: Learning flow factor- ized state space for visual domain generalization

Qi Bi, Jingjun Yi, Hao Zheng, Haolan Zhan, Wei Ji, Yawen Huang, and Yuexiang Li. Dgfamba: Learning flow factor- ized state space for visual domain generalization. InPro- ceedings of the AAAI Conference on Artificial Intelligence, pages 1862–1870, 2025. 2

2025
[6]

Domain generalization by marginal transfer learning.Journal of machine learning re- search, 22(2):1–55, 2021

Gilles Blanchard, Aniket Anand Deshmukh, Urun Dogan, Gyemin Lee, and Clayton Scott. Domain generalization by marginal transfer learning.Journal of machine learning re- search, 22(2):1–55, 2021. 2

2021
[7]

Boffi, Michael S

Nicholas M. Boffi, Michael S. Albergo, and Eric Vanden- Eijnden. How to build a consistency model: Learning flow maps via self-distillation, 2025. 1

2025
[8]

Boffi, Michael S

Nicholas M. Boffi, Michael S. Albergo, and Eric Vanden- Eijnden. Flow map matching with stochastic interpolants: A mathematical framework for consistency models, 2025. 1

2025
[9]

Swad: Domain generalization by seeking flat minima

Junbum Cha, Sanghyuk Chun, Kyungjae Lee, Han-Cheol Cho, Seunghyun Park, Yunsung Lee, and Sungrae Park. Swad: Domain generalization by seeking flat minima. Advances in Neural Information Processing Systems, 34: 22405–22418, 2021. 2

2021
[10]

Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations,
[11]

A simple framework for contrastive learning of visual representations, 2020

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations, 2020. 5

2020
[12]

Jing Du, John Zelek, and Jonathan Li. Weather-aware au- topilot: Domain generalization for point cloud semantic seg- mentation in diverse weather scenarios.ISPRS Journal of Photogrammetry and Remote Sensing, 218:204–219, 2024. 1

2024
[13]

Mitigate the gap: Improving cross-modal alignment in CLIP

Sedigheh Eslami and Gerard de Melo. Mitigate the gap: Improving cross-modal alignment in CLIP. InThe Thir- teenth International Conference on Learning Representa- tions, 2025. 2

2025
[14]

Unbiased met- ric learning: On the utilization of multiple datasets and web images for softening bias

Chen Fang, Ye Xu, and Daniel N Rockmore. Unbiased met- ric learning: On the utilization of multiple datasets and web images for softening bias. InProceedings of the IEEE inter- national conference on computer vision, pages 1657–1664,
[15]

Domain-adversarial training of neural networks.Journal of machine learning research, 17(59):1–35, 2016

Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pas- cal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario March, and Victor Lempitsky. Domain-adversarial training of neural networks.Journal of machine learning research, 17(59):1–35, 2016. 2

2016
[16]

Flow- tok: Flowing seamlessly across text and image tokens, 2025

Ju He, Qihang Yu, Qihao Liu, and Liang-Chieh Chen. Flow- tok: Flowing seamlessly across text and image tokens, 2025. 2

2025
[17]

Momentum contrast for unsupervised visual rep- resentation learning, 2020

Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual rep- resentation learning, 2020. 7

2020
[18]

Denoising diffu- sion probabilistic models, 2020

Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models, 2020. 1, 2

2020
[19]

Self-challenging improves cross-domain generalization

Zeyi Huang, Haohan Wang, Eric P Xing, and Dong Huang. Self-challenging improves cross-domain generalization. In European conference on computer vision, pages 124–140. Springer, 2020. 2

2020
[20]

idag: Invariant dag searching for domain general- ization

Zenan Huang, Haobo Wang, Junbo Zhao, and Nenggan Zheng. idag: Invariant dag searching for domain general- ization. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 19169–19179, 2023. 2

2023
[21]

Out-of-distribution general- ization via risk extrapolation (rex)

David Krueger, Ethan Caballero, Joern-Henrik Jacobsen, Amy Zhang, Jonathan Binas, Dinghuai Zhang, Remi Le Priol, and Aaron Courville. Out-of-distribution general- ization via risk extrapolation (rex). InInternational confer- ence on machine learning, pages 5815–5826. PMLR, 2021. 2

2021
[22]

Sparse mixture- of-experts are domain generalizable learners.arXiv preprint arXiv:2206.04046, 2022

Bo Li, Yifei Shen, Jingkang Yang, Yezhen Wang, Jiawei Ren, Tong Che, Jun Zhang, and Ziwei Liu. Sparse mixture- of-experts are domain generalizable learners.arXiv preprint arXiv:2206.04046, 2022. 2

work page arXiv 2022
[23]

V2x-dgw: Domain generalization for multi-agent percep- tion under adverse weather conditions

Baolu Li, Jinlong Li, Xinyu Liu, Runsheng Xu, Zhengzhong Tu, Jiacheng Guo, Qin Zou, Xiaopeng Li, and Hongkai Yu. V2x-dgw: Domain generalization for multi-agent percep- tion under adverse weather conditions. In2025 IEEE In- ternational Conference on Robotics and Automation (ICRA), pages 974–980. IEEE, 2025. 1

2025
[24]

Hospedales

Da Li, Yongxin Yang, Yi-Zhe Song, and Timothy M. Hospedales. Deeper, broader and artier domain generaliza- tion. InProceedings of the IEEE International Conference on Computer Vision (ICCV), 2017. 4

2017
[25]

Domain generalization for med- ical imaging classification with linear-dependency regular- ization.Advances in neural information processing systems, 33:3118–3129, 2020

Haoliang Li, YuFei Wang, Renjie Wan, Shiqi Wang, Tie- Qiang Li, and Alex Kot. Domain generalization for med- ical imaging classification with linear-dependency regular- ization.Advances in neural information processing systems, 33:3118–3129, 2020. 1

2020
[26]

Mind the gap: Understanding the modality gap in multi-modal contrastive representation learning, 2022

Weixin Liang, Yuhui Zhang, Yongchan Kwon, Serena Ye- ung, and James Zou. Mind the gap: Understanding the modality gap in multi-modal contrastive representation learning, 2022. 1, 2, 6

2022
[27]

Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maxi- milian Nickel, and Matt Le. Flow matching for generative modeling, 2023. 1, 2

2023
[28]

Flowing from words to pixels: A noise-free frame- work for cross-modality evolution

Qihao Liu, Xi Yin, Alan Yuille, Andrew Brown, and Mannat Singh. Flowing from words to pixels: A noise-free frame- work for cross-modality evolution. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 2755–2765, 2025. 2

2025
[29]

Dgmamba: Domain generalization via generalized state space model

Shaocong Long, Qianyu Zhou, Xiangtai Li, Xuequan Lu, Chenhao Ying, Yuan Luo, Lizhuang Ma, and Shuicheng Yan. Dgmamba: Domain generalization via generalized state space model. InProceedings of the 32nd ACM International Conference on Multimedia, pages 3607–3616, 2024. 2

2024
[30]

Unified deep supervised domain adapta- tion and generalization

Saeid Motiian, Marco Piccirilli, Donald A Adjeroh, and Gi- anfranco Doretto. Unified deep supervised domain adapta- tion and generalization. InProceedings of the IEEE inter- national conference on computer vision, pages 5715–5725,
[31]

Domain generalization via invariant fea- ture representation

Krikamol Muandet, David Balduzzi, and Bernhard Sch¨olkopf. Domain generalization via invariant fea- ture representation. InInternational conference on machine learning, pages 10–18. PMLR, 2013. 2

2013
[32]

A Survey on Domain Generalization for Medical Image Analysis,

Ziwei Niu, Shuyi Ouyang, Shiao Xie, Yen-wei Chen, and Lanfen Lin. A survey on domain generalization for medical image analysis.arXiv preprint arXiv:2402.05035, 2024. 1

work page arXiv 2024
[33]

Causality-inspired single- source domain generalization for medical image segmenta- tion.IEEE Transactions on Medical Imaging, 42(4):1095– 1106, 2022

Cheng Ouyang, Chen Chen, Surui Li, Zeju Li, Chen Qin, Wenjia Bai, and Daniel Rueckert. Causality-inspired single- source domain generalization for medical image segmenta- tion.IEEE Transactions on Medical Imaging, 42(4):1095– 1106, 2022. 1

2022
[34]

Learning transferable visual models from natural language supervision, 2021

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. Learning transferable visual models from natural language supervision, 2021. 2

2021
[35]

Hierarchical text-conditional image gener- ation with clip latents, 2022

Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gener- ation with clip latents, 2022. 2

2022
[36]

Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalization

Shiori Sagawa, Pang Wei Koh, Tatsunori B Hashimoto, and Percy Liang. Distributionally robust neural networks for group shifts: On the importance of regularization for worst- case generalization.arXiv preprint arXiv:1911.08731, 2019. 2

work page internal anchor Pith review arXiv 1911
[37]

Domain generalization of 3d semantic segmenta- tion in autonomous driving

Jules Sanchez, Jean-Emmanuel Deschaud, and Franc ¸ois Goulette. Domain generalization of 3d semantic segmenta- tion in autonomous driving. InProceedings of the IEEE/CVF international conference on computer vision, pages 18077– 18087, 2023. 1

2023
[38]

Towards universal representa- tion learning for deep face recognition

Yichun Shi, Xiang Yu, Kihyuk Sohn, Manmohan Chan- draker, and Anil K Jain. Towards universal representa- tion learning for deep face recognition. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6817–6826, 2020. 2

2020
[39]

Deep unsupervised learning using nonequilibrium thermodynamics

Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInternational confer- ence on machine learning, pages 2256–2265. pmlr, 2015. 1

2015
[40]

Score-Based Generative Modeling through Stochastic Differential Equations

Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020. 2

work page internal anchor Pith review Pith/arXiv arXiv 2011
[41]

Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole

Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions, 2021. 1

2021
[42]

Self-distilled vision transformer for domain generalization

Maryam Sultana, Muzammal Naseer, Muhammad Haris Khan, Salman Khan, and Fahad Shahbaz Khan. Self-distilled vision transformer for domain generalization. InProceed- ings of the Asian conference on computer vision, pages 3068–3085, 2022. 2

2022
[43]

Rethinking multi- domain generalization with a general learning objective

Zhaorui Tan, Xi Yang, and Kaizhu Huang. Rethinking multi- domain generalization with a general learning objective. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 23512–23522, 2024. 2

2024
[44]

Repre- sentation learning with contrastive predictive coding, 2019

Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Repre- sentation learning with contrastive predictive coding, 2019. 3, 5

2019
[45]

Deep hashing network for unsupervised domain adaptation

Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep hashing network for unsupervised domain adaptation. InProceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 5018–5027, 2017. 4

2017
[46]

Addressing model vul- nerability to distributional shifts over image transformation sets

Riccardo V olpi and Vittorio Murino. Addressing model vul- nerability to distributional shifts over image transformation sets. InProceedings of the IEEE/CVF International Confer- ence on Computer Vision, pages 7980–7989, 2019. 2

2019
[47]

Deep generative learning via schr ¨odinger bridge,

Gefei Wang, Yuling Jiao, Qian Xu, Yang Wang, and Can Yang. Deep generative learning via schr ¨odinger bridge,
[48]

Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and Philip S. Yu. Generalizing to unseen domains: A survey on domain generalization, 2022. 1

2022
[49]

Sharpness-aware gradient matching for domain generaliza- tion

Pengfei Wang, Zhaoxiang Zhang, Zhen Lei, and Lei Zhang. Sharpness-aware gradient matching for domain generaliza- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, pages 3769–3778,
[50]

Pcl: Proxy-based con- trastive learning for domain generalization

Xufeng Yao, Yang Bai, Xinyun Zhang, Yuechen Zhang, Qi Sun, Ran Chen, Ruiyu Li, and Bei Yu. Pcl: Proxy-based con- trastive learning for domain generalization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7097–7107, 2022. 2

2022
[51]

Explaining and mitigating the modality gap in contrastive multimodal learning.arXiv preprint arXiv:2412.07909, 2024

Can Yaras, Siyi Chen, Peng Wang, and Qing Qu. Explaining and mitigating the modality gap in contrastive multimodal learning.arXiv preprint arXiv:2412.07909, 2024. 2

work page arXiv 2024
[52]

Domain generalization for medical image analysis: A review.Proceedings of the IEEE,

Jee Seok Yoon, Kwanseok Oh, Yooseung Shin, Maciej A Mazurowski, and Heung-Il Suk. Domain generalization for medical image analysis: A review.Proceedings of the IEEE,
[53]

Adaptive risk min- imization: Learning to adapt to domain shift.Advances in Neural Information Processing Systems, 34:23664–23678,

Marvin Zhang, Henrik Marklund, Nikita Dhawan, Abhishek Gupta, Sergey Levine, and Chelsea Finn. Adaptive risk min- imization: Learning to adapt to domain shift.Advances in Neural Information Processing Systems, 34:23664–23678,
[54]

Learning a cross-modal schr¨odinger bridge for visual domain generalization

Hao Zheng, Jingjun Yi, Qi Bi, Huimin Huang, Haolan Zhan, Yawen Huang, Yuexiang Li, Xian Wu, and Yefeng Zheng. Learning a cross-modal schr¨odinger bridge for visual domain generalization. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. 2

2025
[55]

Deep domain-adversarial image generation for do- main generalisation

Kaiyang Zhou, Yongxin Yang, Timothy Hospedales, and Tao Xiang. Deep domain-adversarial image generation for do- main generalisation. InProceedings of the AAAI conference on artificial intelligence, pages 13025–13032, 2020. 2

2020
[56]

arXiv preprint arXiv:2104.02008 , year=

Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xi- ang. Domain generalization with mixstyle.arXiv preprint arXiv:2104.02008, 2021. 2

work page arXiv 2021
[57]

Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, page 1–20, 2022

Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, page 1–20, 2022. 1, 2

2022
[58]

Mixstyle neural networks for domain generalization and adaptation.International Journal of Computer Vision, 132 (3):822–836, 2024

Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xiang. Mixstyle neural networks for domain generalization and adaptation.International Journal of Computer Vision, 132 (3):822–836, 2024. 2 CrossFlowDG: Bridging the Modality Gap with Cross-modal Flow Matching for Domain Generalization Supplementary Material A. Textual Domain Bank Entries Table S1 enumerates th...

2024