Yong Rui
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Yong Rui has a total of 105 co-authors. Here are the top co-authors by number of shared publications:
- Xin Geng
School of Computer Science and Engineering, Southeast University, Nanjing, China and Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, Nanjing, China
6 shared publications - Jin Yuan
Lenovo Ltd, Beijing, China
5 shared publications - Jianping Fan
AI Lab at Lenovo Research, Beijing, China
4 shared publications - Shuqiang Jiang
University of Chinese Academy of Sciences, Beijing, China
4 shared publications - Weiqing Min
Key Laboratory of Intelligent Information Processing, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China
4 shared publications - Zhongchao Shi
AI Lab, Lenovo Research, Beijing, China
4 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1322-1331
Physics Attention: A Physics-Learned Generative World Model with Disentangled Latent Physical Factors for Fluid
Generative world models have demonstrated strong capability in modeling complex environment dynamics. However, existing approaches mainly rely on data pattern matching or physics-informed generation with externally given physical priors, failing to learn the intrinsic and coupled physical laws. Those challenges are particularly severe in fluid environments, where governing physical laws are implicit and jointly determined by multiple related physical factors. To solve these challenges, we propose the Physics Attention World Model (PAWM), a physics-learned generative world model that autonomously learns and decorrelated informs latent physics from raw fluid observations. Specifically, PAWM comprises two core modules: i) the Physics Attention module, which learns disentangled latent physics and achieves physics-based consistent constraints, and ii) the Decorrelated Physics-informed Prediction module, which first conducts decorrelation reweighting to quantify the contribution of different physical factors in video generation with theoretical support, then performs state transition and prediction conditioned on these learned and decorrelated latent physics. Extensive experiments on three fluid benchmarks demonstrate that PAWM significantly outperforms existing generative world models with respect to accurate and physically consistent prediction across both simulated and real-world fluid scenarios, as well as maintaining strong generalization across diverse fluid properties.
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1322-1331
Open AccessPhysics Attention: A Physics-Learned Generative World Model with Disentangled Latent Physical Factors for Fluid
Generative world models have demonstrated strong capability in modeling complex environment dynamics. However, existing approaches mainly rely on data pattern matching or physics-informed generation with externally given physical priors, failing to learn the intrinsic and coupled physical laws. Those challenges are particularly severe in fluid environments, where governing physical laws are implicit and jointly determined by multiple related physical factors. To solve these challenges, we propose the Physics Attention World Model (PAWM), a physics-learned generative world model that autonomously learns and decorrelated informs latent physics from raw fluid observations. Specifically, PAWM comprises two core modules: i) the Physics Attention module, which learns disentangled latent physics and achieves physics-based consistent constraints, and ii) the Decorrelated Physics-informed Prediction module, which first conducts decorrelation reweighting to quantify the contribution of different physical factors in video generation with theoretical support, then performs state transition and prediction conditioned on these learned and decorrelated latent physics. Extensive experiments on three fluid benchmarks demonstrate that PAWM significantly outperforms existing generative world models with respect to accurate and physically consistent prediction across both simulated and real-world fluid scenarios, as well as maintaining strong generalization across diverse fluid properties.
- Open Access
- 8 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (6) , 1-24
A Bootstrap Pipeline for Chat-Based Image Retrieval with Effective Question Generation
Chat-based image retrieval uses Large Language Models (LLM) to guide user input to enable more specific and precise search results, where LLM can enhance this process by asking user retrieval-oriented questions eliciting additional details about the target image. Despite the potential of this approach, no specialized Questioner model has been developed for this task due to the following significant challenges: (a) the difficulty of determining the optimal questions to ask; (b) the lack of a suitable protocol for fair model comparison; and (c) the notable scarcity of dialog-to-image retrieval data. To address these challenges, two fundamental principles are developed in this article to ensure the simplicity and effectiveness of the generated questions while enabling a fair comparison and accurate estimation of data quality and model performance. A bootstrap training methodology is introduced to collect retrieval-oriented dialog data and concurrently train the Questioner and the image Retriever. Under a fair comparison protocol, our extensive experiments have demonstrated that our proposed method can not only address the critical data gap but also achieve state-of-the-art results, which substantially surpass GPT-4o and GPT-4-Turbo through the fine-tuning of an 8B model.
- 8 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (6) , 1-24
Open AccessA Bootstrap Pipeline for Chat-Based Image Retrieval with Effective Question Generation
Chat-based image retrieval uses Large Language Models (LLM) to guide user input to enable more specific and precise search results, where LLM can enhance this process by asking user retrieval-oriented questions eliciting additional details about the target image. Despite the potential of this approach, no specialized Questioner model has been developed for this task due to the following significant challenges: (a) the difficulty of determining the optimal questions to ask; (b) the lack of a suitable protocol for fair model comparison; and (c) the notable scarcity of dialog-to-image retrieval data. To address these challenges, two fundamental principles are developed in this article to ensure the simplicity and effectiveness of the generated questions while enabling a fair comparison and accurate estimation of data quality and model performance. A bootstrap training methodology is introduced to collect retrieval-oriented dialog data and concurrently train the Questioner and the image Retriever. Under a fair comparison protocol, our extensive experiments have demonstrated that our proposed method can not only address the critical data gap but also achieve state-of-the-art results, which substantially surpass GPT-4o and GPT-4-Turbo through the fine-tuning of an 8B model.
- Open Access
Thinking inside the Convolution for Image Inpainting: Reconstructing Texture via Structure under Global and Local Side
Image inpainting has earned substantial progress, owing to the encoder-and-decoder pipeline, which is benefited from the Convolutional Neural Networks (CNNs) with convolutional downsampling to inpaint the masked regions semantically from the known regions within the encoder, coupled with an upsampling process from the decoder for final inpainting output. Recent studies intuitively identify the high-frequency structure and low-frequency texture to be extracted by CNNs from the encoder, and subsequently for a desirable upsampling recovery. However, the existing arts inevitably overlook the information loss for both structure and texture feature maps during the convolutional downsampling process, hence suffer from a non-ideal upsampling output. In this paper, we systematically answer whether and how the structure and texture feature map can mutually help to alleviate the information loss during the convolutional downsampling. Given the structure and texture feature maps, we adopt the statistical normalization and denormalization strategy for the reconstruction guidance during the convolutional downsampling process. The extensive experimental results validate its advantages to the state-of-the-arts over the images from low-to-high resolutions including 256*256 and 512*512, especially holds by substituting all the encoders by ours. Our code is available at https://github.com/htyjers/ConvInpaint-TSGL
Open AccessThinking inside the Convolution for Image Inpainting: Reconstructing Texture via Structure under Global and Local Side
Image inpainting has earned substantial progress, owing to the encoder-and-decoder pipeline, which is benefited from the Convolutional Neural Networks (CNNs) with convolutional downsampling to inpaint the masked regions semantically from the known regions within the encoder, coupled with an upsampling process from the decoder for final inpainting output. Recent studies intuitively identify the high-frequency structure and low-frequency texture to be extracted by CNNs from the encoder, and subsequently for a desirable upsampling recovery. However, the existing arts inevitably overlook the information loss for both structure and texture feature maps during the convolutional downsampling process, hence suffer from a non-ideal upsampling output. In this paper, we systematically answer whether and how the structure and texture feature map can mutually help to alleviate the information loss during the convolutional downsampling. Given the structure and texture feature maps, we adopt the statistical normalization and denormalization strategy for the reconstruction guidance during the convolutional downsampling process. The extensive experimental results validate its advantages to the state-of-the-arts over the images from low-to-high resolutions including 256*256 and 512*512, especially holds by substituting all the encoders by ours. Our code is available at https://github.com/htyjers/ConvInpaint-TSGL
- Top Cited Papers
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-12
LMAgent: A Large-scale Multimodal Agents Society for Multi-user Simulation
The believable simulation of multi-user behavior is crucial for understanding complex social systems. Recently, large language models (LLMs)-based AI agents have made signifi cant progress, enabling them to achieve human-like intelligence across various tasks. However, real human societies are often dynamic and complex, involving numerous individuals engaging in multimodal interactions. In this paper, taking e-commerce scenarios as an example, we present LMAgent, a very large scale and multimodal agents society based on multimodal LLMs. In LMAgent, besides freely chatting with friends, the agents can autonomously browse, purchase, and review products, even perform live streaming e-commerce. To simulate this complex system, we introduce a self-consistency prompting mechanism to augment agents' multimodal capabilities, resulting in significantly improved decision-making performance over the existing multi agent system. Moreover, we propose a fast memory mechanism combined with the small-world model to enhance system ef f iciency, which supports more than 10,000 agent simulations in a society. Experiments on agents' behavior show that these agents achieve comparable performance to humans in behavioral indicators. Furthermore, compared with the existing LLMs-based multi-agent system, more different and valuable phenomena are exhibited, such as herd behavior, which demonstrates the potential of LMAgent in credible large-scale social behavior simulations.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-12
Top Cited PapersLMAgent: A Large-scale Multimodal Agents Society for Multi-user Simulation
The believable simulation of multi-user behavior is crucial for understanding complex social systems. Recently, large language models (LLMs)-based AI agents have made signifi cant progress, enabling them to achieve human-like intelligence across various tasks. However, real human societies are often dynamic and complex, involving numerous individuals engaging in multimodal interactions. In this paper, taking e-commerce scenarios as an example, we present LMAgent, a very large scale and multimodal agents society based on multimodal LLMs. In LMAgent, besides freely chatting with friends, the agents can autonomously browse, purchase, and review products, even perform live streaming e-commerce. To simulate this complex system, we introduce a self-consistency prompting mechanism to augment agents' multimodal capabilities, resulting in significantly improved decision-making performance over the existing multi agent system. Moreover, we propose a fast memory mechanism combined with the small-world model to enhance system ef f iciency, which supports more than 10,000 agent simulations in a society. Experiments on agents' behavior show that these agents achieve comparable performance to humans in behavioral indicators. Furthermore, compared with the existing LLMs-based multi-agent system, more different and valuable phenomena are exhibited, such as herd behavior, which demonstrates the potential of LMAgent in credible large-scale social behavior simulations.
- 11 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (3) , 3082-3099
Semi-Supervised Unconstrained Head Pose Estimation in the Wild
Existing research on unconstrained in-the-wild head pose estimation suffers from the flaws of its datasets, which consist of either numerous samples by non-realistic synthesis or constrained collection, or small-scale natural images yet with plausible manual annotations. This makes fully-supervised solutions compromised due to the reliance on generous labels. To alleviate it, we propose the first semi-supervised unconstrained head pose estimation method SemiUHPE, which can leverage abundant easily available unlabeled head images. Technically, we choose semi-supervised rotation regression and adapt it to the error-sensitive and label-scarce problem of unconstrained head pose. Our method is based on the observation that the aspect-ratio invariant cropping of wild heads is superior to previous landmark-based affine alignment given that landmarks of unconstrained human heads are usually unavailable, especially for underexplored non-frontal heads. Instead of using a pre-fixed threshold to filter out pseudo labeled heads, we propose dynamic entropy based filtering to adaptively remove unlabeled outliers as training progresses by updating the threshold in multiple stages. We then revisit the design of weak-strong augmentations and improve it by devising two novel head-oriented strong augmentations, termed pose-irrelevant cut-occlusion and pose-altering rotation consistency respectively. Extensive experiments and ablation studies show that SemiUHPE outperforms its counterparts greatly on public benchmarks under both the front-range and full-range settings. Furthermore, our proposed method is also beneficial for solving other closely related problems, including generic object rotation regression and 3D head reconstruction, demonstrating good versatility and extensibility.
- 11 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (3) , 3082-3099
Semi-Supervised Unconstrained Head Pose Estimation in the Wild
Existing research on unconstrained in-the-wild head pose estimation suffers from the flaws of its datasets, which consist of either numerous samples by non-realistic synthesis or constrained collection, or small-scale natural images yet with plausible manual annotations. This makes fully-supervised solutions compromised due to the reliance on generous labels. To alleviate it, we propose the first semi-supervised unconstrained head pose estimation method SemiUHPE, which can leverage abundant easily available unlabeled head images. Technically, we choose semi-supervised rotation regression and adapt it to the error-sensitive and label-scarce problem of unconstrained head pose. Our method is based on the observation that the aspect-ratio invariant cropping of wild heads is superior to previous landmark-based affine alignment given that landmarks of unconstrained human heads are usually unavailable, especially for underexplored non-frontal heads. Instead of using a pre-fixed threshold to filter out pseudo labeled heads, we propose dynamic entropy based filtering to adaptively remove unlabeled outliers as training progresses by updating the threshold in multiple stages. We then revisit the design of weak-strong augmentations and improve it by devising two novel head-oriented strong augmentations, termed pose-irrelevant cut-occlusion and pose-altering rotation consistency respectively. Extensive experiments and ablation studies show that SemiUHPE outperforms its counterparts greatly on public benchmarks under both the front-range and full-range settings. Furthermore, our proposed method is also beneficial for solving other closely related problems, including generic object rotation regression and 3D head reconstruction, demonstrating good versatility and extensibility.
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3280-3289
Building Embodied EvoAgent: A Brain-inspired Paradigm for Bridging Multimodal Large Models and World Models
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 3280-3289
Open AccessBuilding Embodied EvoAgent: A Brain-inspired Paradigm for Bridging Multimodal Large Models and World Models
- Open Access
- 25 September 2025
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 19 (12) , 1-10
Collective domain adversarial learning for unsupervised domain adaptation
- 25 September 2025
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 19 (12) , 1-10
Open AccessCollective domain adversarial learning for unsupervised domain adaptation
- 25 September 2025
- Published by Wiley in Land Degradation & Development
- Vol. 37 (5) , 1673-1691
Piping Erosion in Semi‐Arid Loess Hillslope: The Influence of Soil Characteristics
Piping erosion, driven by subsurface flow, has generally received less attention than soil erosion caused by overland flow. This oversight contributes to the growing global problem of soil degradation in various regions. Piping erosion can pose significant geohazards in areas covered by loess, such as the Iranian Loess Plateau (ILP), where it negatively impacts agricultural productivity and human properties. Consequently, identifying the factors influencing this erosion process is essential for developing effective control measures. Therefore, this study aimed to examine the role of soil characteristics and slope positions in the selected area of the ILP prone to piping erosion. In order to achieve this goal, six soil profiles were excavated both inside and outside pipe collapses (PCs) and at various positions on the hillslope (shoulder, backslope, and toeslope). The soil profiles were described following standard pedological methods (the Soil Survey Manual). A total of 28 soil samples were analyzed in the laboratory. Apart from cluster analysis, a comparison of soil properties was done: (1) at various slope positions, and (2) inside and outside PCs. The silt content of the analyzed soils ranges from 57% to 84%, predominantly consisting of medium‐sized silt particles, generally making these soils vulnerable to soil erosion. The sodium (Na + ) content tends to decrease in soil profiles inside the PCs, while the exchangeable sodium percentage (ESP) generally increases toward deeper soil layers. This condition may promote preferential flow, particularly in the lower horizons, which acts as a primary driving force for the initiation of piping erosion. Cluster analysis revealed that high content of Na + , Mg 2+ , and high ESP values are the most significant contributors to piping erosion. Notably, the deepest soil horizons in the pits inside the PCs and across different hillslope positions were grouped into the same cluster, indicating that the C horizons of profiles inside the PCs are the most vulnerable to piping erosion. Variations in vertical soil structure, both within and outside the PCs at different slope positions, suggest that weak soil structure may significantly influence the initiation and enlargement of pipes. The deepest soil horizons of soil profiles inside the PCs have the highest ESP values, indicating their dispersive characteristics. This study has shown that, in loess‐derived soils under semi‐arid conditions, soil properties have a greater impact on piping erosion than hillslope positions. In summary, the study area is characterized by erodible soils, with all hillslope positions being highly susceptible to piping. This research enhances our understanding of piping erosion in loess‐covered regions under semi‐arid conditions and provides a crucial step toward developing effective soil erosion control measures.
- 25 September 2025
- Published by Wiley in Land Degradation & Development
- Vol. 37 (5) , 1673-1691
Piping Erosion in Semi‐Arid Loess Hillslope: The Influence of Soil Characteristics
Piping erosion, driven by subsurface flow, has generally received less attention than soil erosion caused by overland flow. This oversight contributes to the growing global problem of soil degradation in various regions. Piping erosion can pose significant geohazards in areas covered by loess, such as the Iranian Loess Plateau (ILP), where it negatively impacts agricultural productivity and human properties. Consequently, identifying the factors influencing this erosion process is essential for developing effective control measures. Therefore, this study aimed to examine the role of soil characteristics and slope positions in the selected area of the ILP prone to piping erosion. In order to achieve this goal, six soil profiles were excavated both inside and outside pipe collapses (PCs) and at various positions on the hillslope (shoulder, backslope, and toeslope). The soil profiles were described following standard pedological methods (the Soil Survey Manual). A total of 28 soil samples were analyzed in the laboratory. Apart from cluster analysis, a comparison of soil properties was done: (1) at various slope positions, and (2) inside and outside PCs. The silt content of the analyzed soils ranges from 57% to 84%, predominantly consisting of medium‐sized silt particles, generally making these soils vulnerable to soil erosion. The sodium (Na + ) content tends to decrease in soil profiles inside the PCs, while the exchangeable sodium percentage (ESP) generally increases toward deeper soil layers. This condition may promote preferential flow, particularly in the lower horizons, which acts as a primary driving force for the initiation of piping erosion. Cluster analysis revealed that high content of Na + , Mg 2+ , and high ESP values are the most significant contributors to piping erosion. Notably, the deepest soil horizons in the pits inside the PCs and across different hillslope positions were grouped into the same cluster, indicating that the C horizons of profiles inside the PCs are the most vulnerable to piping erosion. Variations in vertical soil structure, both within and outside the PCs at different slope positions, suggest that weak soil structure may significantly influence the initiation and enlargement of pipes. The deepest soil horizons of soil profiles inside the PCs have the highest ESP values, indicating their dispersive characteristics. This study has shown that, in loess‐derived soils under semi‐arid conditions, soil properties have a greater impact on piping erosion than hillslope positions. In summary, the study area is characterized by erodible soils, with all hillslope positions being highly susceptible to piping. This research enhances our understanding of piping erosion in loess‐covered regions under semi‐arid conditions and provides a crucial step toward developing effective soil erosion control measures.
- Open Access
- 19 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (7) , 1-28
Multimodal Food Learning
Food-centered study has received more attention in the multimedia community for its profound impact on our survival, nutrition and health, pleasure, and enjoyment. Our experience of food is typically multi-sensory: We see food objects, smell its odors, taste its flavors, feel its texture, and hear sounds when chewing. Therefore, multimodal food learning is vital in food-centered study, which aims to relate information from multiple food modalities to support various multimedia tasks, ranging from recognition, retrieval, generation, recommendation, and interaction, enabling applications in different fields like healthcare and agriculture. However, there is no surveys on this topic to our knowledge. To fill this gap, this article formalizes multimodal food learning and comprehensively surveys its typical tasks, technical achievements, existing datasets, and applications to provide the blueprint with researchers and practitioners. Based on the current state of the art, we identify both open research issues and promising research directions, such as multimodal food learning benchmark construction, multimodal food foundation model construction, and multimodality diet estimation. We also point out that closer cooperation from researchers between multimedia and food science can handle some existing challenges and meanwhile open up more new opportunities to advance the fast development of multimodal food learning. This is the first comprehensive survey in this topic and we anticipate about 170 reviewed research articles can benefit academia and industry in this community and beyond.
- 19 July 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (7) , 1-28
Open AccessMultimodal Food Learning
Food-centered study has received more attention in the multimedia community for its profound impact on our survival, nutrition and health, pleasure, and enjoyment. Our experience of food is typically multi-sensory: We see food objects, smell its odors, taste its flavors, feel its texture, and hear sounds when chewing. Therefore, multimodal food learning is vital in food-centered study, which aims to relate information from multiple food modalities to support various multimedia tasks, ranging from recognition, retrieval, generation, recommendation, and interaction, enabling applications in different fields like healthcare and agriculture. However, there is no surveys on this topic to our knowledge. To fill this gap, this article formalizes multimodal food learning and comprehensively surveys its typical tasks, technical achievements, existing datasets, and applications to provide the blueprint with researchers and practitioners. Based on the current state of the art, we identify both open research issues and promising research directions, such as multimodal food learning benchmark construction, multimodal food foundation model construction, and multimodality diet estimation. We also point out that closer cooperation from researchers between multimedia and food science can handle some existing challenges and meanwhile open up more new opportunities to advance the fast development of multimodal food learning. This is the first comprehensive survey in this topic and we anticipate about 170 reviewed research articles can benefit academia and industry in this community and beyond.
- 4 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27 (15209210) , 6648-6660
DomainVerse: A Benchmark Towards Real-World Distribution Shifts for Training-Free Adaptive Domain Generalization
Traditional cross-domain tasks, including unsupervised domain adaptation (UDA), domain generalization (DG) and test-time adaptation (TTA), rely heavily on the training model by source domain data whether for specific or arbitrary target domains. With the recent advance of vision-language models (VLMs), recognized as natural source models that can be transferred to various downstream tasks without any parameter training, we propose a novel cross-domain task directly combining the strengths of both UDA and DG, named Training-Free Adaptive Domain Generalization (TF-ADG). However, current cross-domain datasets have many limitations, such as unrealistic domains, unclear domain definitions, and the inability to fine-grained domain decomposition, which hinder the real-world application of current cross-domain models due to the lack of accurate and fair evaluation of fine-grained realistic domains. These insights motivate us to establish a novel realistic benchmark for TF-ADG. Benefiting from the introduced hierarchical definition of domain shifts, our proposed dataset DomainVerse addresses these issues by providing about 0.5 million images from 390 realistic, hierarchical, and balanced domains, allowing for decomposition across multiple domains within each image. With the help of the constructed DomainVerse and VLMs, we further propose two algorithms called Domain CLIP and Domain++ CLIP for training-free adaptive domain generalization. Extensive and comprehensive experiments demonstrate the significance of the dataset and the effectiveness of the proposed methods.
- 4 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27 (15209210) , 6648-6660
DomainVerse: A Benchmark Towards Real-World Distribution Shifts for Training-Free Adaptive Domain Generalization
Traditional cross-domain tasks, including unsupervised domain adaptation (UDA), domain generalization (DG) and test-time adaptation (TTA), rely heavily on the training model by source domain data whether for specific or arbitrary target domains. With the recent advance of vision-language models (VLMs), recognized as natural source models that can be transferred to various downstream tasks without any parameter training, we propose a novel cross-domain task directly combining the strengths of both UDA and DG, named Training-Free Adaptive Domain Generalization (TF-ADG). However, current cross-domain datasets have many limitations, such as unrealistic domains, unclear domain definitions, and the inability to fine-grained domain decomposition, which hinder the real-world application of current cross-domain models due to the lack of accurate and fair evaluation of fine-grained realistic domains. These insights motivate us to establish a novel realistic benchmark for TF-ADG. Benefiting from the introduced hierarchical definition of domain shifts, our proposed dataset DomainVerse addresses these issues by providing about 0.5 million images from 390 realistic, hierarchical, and balanced domains, allowing for decomposition across multiple domains within each image. With the help of the constructed DomainVerse and VLMs, we further propose two algorithms called Domain CLIP and Domain++ CLIP for training-free adaptive domain generalization. Extensive and comprehensive experiments demonstrate the significance of the dataset and the effectiveness of the proposed methods.