The Telephone Game: Evaluating Semantic Drift in Unified Models

Employing a single, unified model (UM) for both visual understanding (image-to-text: I2T) and and visual generation (text-to-image: T2I) has opened a new direction in Visual Language Model (VLM) research. While UMs can also support broader unimodal tasks (e.g., text-to-text, image-to-image), we focus on the core cross-modal pair T2I and I2T, as consistency between understanding and generation is critical for downstream use. Existing evaluations consider these capabilities in isolation: FID and GenEval for T2I, and benchmarks such as MME, MMBench for I2T. These single-pass metrics do not reveal whether a model that understands a concept can also render it, nor whether meaning is preserved when cycling between image and text modalities. To address this, we introduce the Unified Consistency Framework for Unified Models (UCF-UM), a cyclic evaluation protocol that alternates I2T and T2I over multiple generations to quantify semantic drift. UCF formulates 3 metrics: (i) Mean Cumulative Drift (MCD), an embedding-based measure of overall semantic loss; (ii) Semantic Drift Rate (SDR), that summarizes semantic decay rate; and (iii) Multi-Generation GenEval (MGG), an object-level compliance score extending GenEval. To assess generalization beyond COCO, which is widely used in training; we create a new benchmark ND400, sampled from NoCaps and DOCCI and evaluate on seven recent models. UCF-UM reveals substantial variation in cross-modal stability: some models like BAGEL maintain semantics over many alternations, whereas others like Vila-u drift quickly despite strong single-pass scores. Our results highlight cyclic consistency as a necessary complement to standard I2T and T2I evaluations, and provide practical metrics to consistently assess unified model's cross-modal stability and strength of their shared representations. Code: https://github.com/mollahsabbir/Semantic-Drift-in-Unified-Models

Self-adaptive Dataset Construction for Real-World Multimodal Safety Scenarios

Multimodal large language models (MLLMs) are rapidly evolving, presenting increasingly complex safety challenges. However, current dataset construction methods, which are risk-oriented, fail to cover the growing complexity of real-world multimodal safety scenarios (RMS). And due to the lack of a unified evaluation metric, their overall effectiveness remains unproven. This paper introduces a novel image-oriented self-adaptive dataset construction method for RMS, which starts with images and end constructing paired text and guidance responses. Using the image-oriented method, we automatically generate an RMS dataset comprising 35k image-text pairs with guidance responses. Additionally, we introduce a standardized safety dataset evaluation metric: fine-tuning a safety judge model and evaluating its capabilities on other safety datasets.Extensive experiments on various tasks demonstrate the effectiveness of the proposed image-oriented pipeline. The results confirm the scalability and effectiveness of the image-oriented approach, offering a new perspective for the construction of real-world multimodal safety datasets.

AnomalyLMM: Bridging Generative Knowledge and Discriminative Retrieval for Text-Based Person Anomaly Search

With growing public safety demands, text-based person anomaly search has emerged as a critical task, aiming to retrieve individuals with abnormal behaviors via natural language descriptions. Unlike conventional person search, this task presents two unique challenges: (1) fine-grained cross-modal alignment between textual anomalies and visual behaviors, and (2) anomaly recognition under sparse real-world samples. While Large Multi-modal Models (LMMs) excel in multi-modal understanding, their potential for fine-grained anomaly retrieval remains underexplored, hindered by: (1) a domain gap between generative knowledge and discriminative retrieval, and (2) the absence of efficient adaptation strategies for deployment. In this work, we propose AnomalyLMM, the first framework that harnesses LMMs for text-based person anomaly search. Our key contributions are: (1) A novel coarse-to-fine pipeline integrating LMMs to bridge generative world knowledge with retrieval-centric anomaly detection; (2) A training-free adaptation cookbook featuring masked cross-modal prompting, behavioral saliency prediction, and knowledge-aware re-ranking, enabling zero-shot focus on subtle anomaly cues. As the first study to explore LMMs for this task, we conduct a rigorous evaluation on the PAB dataset, the only publicly available benchmark for text-based person anomaly search, with its curated real-world anomalies covering diverse scenarios (e.g., falling, collision, and being hit). Experiments show the effectiveness of the proposed method, surpassing the competitive baseline by +0.96% Recall@1 accuracy. Notably, our method reveals interpretable alignment between textual anomalies and visual behaviors, validated via qualitative analysis. Our code and models will be released for future research.

SRWToolkit: An Open Source Wizard of Oz Toolkit to Create Social Robotic Avatars

We present SRWToolkit, an open-source Wizard of Oz toolkit designed to facilitate the rapid prototyping of social robotic avatars powered by local large language models (LLMs). Our web-based toolkit enables multimodal interaction through text input, button-activated speech, and wake-word command. The toolkit offers real-time configuration of avatar appearance, behavior, language, and voice via an intuitive control panel. In contrast to prior works that rely on cloud-based LLM services, SRWToolkit emphasizes modularity and ensures on-device functionality through local LLM inference. In our small-scale user study ($n=11$), participants created and interacted with diverse robotic roles (hospital receptionist, mathematics teacher, and driving assistant), which demonstrated positive outcomes in the toolkit's usability, trust, and user experience. The toolkit enables rapid and efficient development of robot characters customized to researchers' needs, supporting scalable research in human-robot interaction.

From Editor to Dense Geometry Estimator

Leveraging visual priors from pre-trained text-to-image (T2I) generative models has shown success in dense prediction. However, dense prediction is inherently an image-to-image task, suggesting that image editing models, rather than T2I generative models, may be a more suitable foundation for fine-tuning. Motivated by this, we conduct a systematic analysis of the fine-tuning behaviors of both editors and generators for dense geometry estimation. Our findings show that editing models possess inherent structural priors, which enable them to converge more stably by ``refining" their innate features, and ultimately achieve higher performance than their generative counterparts. Based on these findings, we introduce \textbf{FE2E}, a framework that pioneeringly adapts an advanced editing model based on Diffusion Transformer (DiT) architecture for dense geometry prediction. Specifically, to tailor the editor for this deterministic task, we reformulate the editor's original flow matching loss into the ``consistent velocity" training objective. And we use logarithmic quantization to resolve the precision conflict between the editor's native BFloat16 format and the high precision demand of our tasks. Additionally, we leverage the DiT's global attention for a cost-free joint estimation of depth and normals in a single forward pass, enabling their supervisory signals to mutually enhance each other. Without scaling up the training data, FE2E achieves impressive performance improvements in zero-shot monocular depth and normal estimation across multiple datasets. Notably, it achieves over 35\% performance gains on the ETH3D dataset and outperforms the DepthAnything series, which is trained on 100$\times$ data. The project page can be accessed \href{https://amap-ml.github.io/FE2E/}{here}.

Temporal Interest-Driven Multimodal Personalized Content Generation

With the dynamic evolution of user interests and the increasing multimodal demands in internet applications, personalized content generation strategies based on static interest preferences struggle to meet practical application requirements. The proposed TIMGen (Temporal Interest-driven Multimodal Generation) model addresses this challenge by modeling the long-term temporal evolution of users' interests and capturing dynamic interest representations with strong temporal dependencies. This model also supports the fusion of multimodal features, such as text, images, video, and audio, and delivers customized content based on multimodal preferences. TIMGen jointly learns temporal dependencies and modal preferences to obtain a unified interest representation, which it then generates to meet users' personalized content needs. TIMGen overcomes the shortcomings of personalized content recommendation methods based on static preferences, enabling flexible and dynamic modeling of users' multimodal interests, better understanding and capturing their interests and preferences. It can be extended to a variety of practical application scenarios, including e-commerce, advertising, online education, and precision medicine, providing insights for future research.

Noisy Label Refinement with Semantically Reliable Synthetic Images

Semantic noise in image classification datasets, where visually similar categories are frequently mislabeled, poses a significant challenge to conventional supervised learning approaches. In this paper, we explore the potential of using synthetic images generated by advanced text-to-image models to address this issue. Although these high-quality synthetic images come with reliable labels, their direct application in training is limited by domain gaps and diversity constraints. Unlike conventional approaches, we propose a novel method that leverages synthetic images as reliable reference points to identify and correct mislabeled samples in noisy datasets. Extensive experiments across multiple benchmark datasets show that our approach significantly improves classification accuracy under various noise conditions, especially in challenging scenarios with semantic label noise. Additionally, since our method is orthogonal to existing noise-robust learning techniques, when combined with state-of-the-art noise-robust training methods, it achieves superior performance, improving accuracy by 30% on CIFAR-10 and by 11% on CIFAR-100 under 70% semantic noise, and by 24% on ImageNet-100 under real-world noise conditions.

TauGenNet: Plasma-Driven Tau PET Image Synthesis via Text-Guided 3D Diffusion Models

Accurate quantification of tau pathology via tau positron emission tomography (PET) scan is crucial for diagnosing and monitoring Alzheimer's disease (AD). However, the high cost and limited availability of tau PET restrict its widespread use. In contrast, structural magnetic resonance imaging (MRI) and plasma-based biomarkers provide non-invasive and widely available complementary information related to brain anatomy and disease progression. In this work, we propose a text-guided 3D diffusion model for 3D tau PET image synthesis, leveraging multimodal conditions from both structural MRI and plasma measurement. Specifically, the textual prompt is from the plasma p-tau217 measurement, which is a key indicator of AD progression, while MRI provides anatomical structure constraints. The proposed framework is trained and evaluated using clinical AV1451 tau PET data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Experimental results demonstrate that our approach can generate realistic, clinically meaningful 3D tau PET across a range of disease stages. The proposed framework can help perform tau PET data augmentation under different settings, provide a non-invasive, cost-effective alternative for visualizing tau pathology, and support the simulation of disease progression under varying plasma biomarker levels and cognitive conditions.

MEPG:Multi-Expert Planning and Generation for Compositionally-Rich Image Generation

Text-to-image diffusion models have achieved remarkable image quality, but they still struggle with complex, multiele ment prompts, and limited stylistic diversity. To address these limitations, we propose a Multi-Expert Planning and Gen eration Framework (MEPG) that synergistically integrates position- and style-aware large language models (LLMs) with spatial-semantic expert modules. The framework comprises two core components: (1) a Position-Style-Aware (PSA) module that utilizes a supervised fine-tuned LLM to decom pose input prompts into precise spatial coordinates and style encoded semantic instructions; and (2) a Multi-Expert Dif fusion (MED) module that implements cross-region genera tion through dynamic expert routing across both local regions and global areas. During the generation process for each lo cal region, specialized models (e.g., realism experts, styliza tion specialists) are selectively activated for each spatial par tition via attention-based gating mechanisms. The architec ture supports lightweight integration and replacement of ex pert models, providing strong extensibility. Additionally, an interactive interface enables real-time spatial layout editing and per-region style selection from a portfolio of experts. Ex periments show that MEPG significantly outperforms base line models with the same backbone in both image quality and style diversity.

SMooGPT: Stylized Motion Generation using Large Language Models

Stylized motion generation is actively studied in computer graphics, especially benefiting from the rapid advances in diffusion models. The goal of this task is to produce a novel motion respecting both the motion content and the desired motion style, e.g., ``walking in a loop like a Monkey''. Existing research attempts to address this problem via motion style transfer or conditional motion generation. They typically embed the motion style into a latent space and guide the motion implicitly in a latent space as well. Despite the progress, their methods suffer from low interpretability and control, limited generalization to new styles, and fail to produce motions other than ``walking'' due to the strong bias in the public stylization dataset. In this paper, we propose to solve the stylized motion generation problem from a new perspective of reasoning-composition-generation, based on our observations: i) human motion can often be effectively described using natural language in a body-part centric manner, ii) LLMs exhibit a strong ability to understand and reason about human motion, and iii) human motion has an inherently compositional nature, facilitating the new motion content or style generation via effective recomposing. We thus propose utilizing body-part text space as an intermediate representation, and present SMooGPT, a fine-tuned LLM, acting as a reasoner, composer, and generator when generating the desired stylized motion. Our method executes in the body-part text space with much higher interpretability, enabling fine-grained motion control, effectively resolving potential conflicts between motion content and style, and generalizes well to new styles thanks to the open-vocabulary ability of LLMs. Comprehensive experiments and evaluations, and a user perceptual study, demonstrate the effectiveness of our approach, especially under the pure text-driven stylized motion generation.