Yiming Shi

Yujia Wu, Yiming Shi, Jiwei Wei et al.

Personalized text-to-image generation has gained significant attention for its capability to generate high-fidelity portraits of specific identities conditioned on user-defined prompts. Existing methods typically involve test-time fine-tuning or incorporating an additional pre-trained branch. However, these approaches struggle to simultaneously address efficiency, identity fidelity, and the preservation of the model's original generative capabilities. In this paper, we propose DiffLoRA, an efficient method that leverages the diffusion model as a hypernetwork to predict personalized Low-Rank Adaptation (LoRA) weights based on the reference images. By incorporating these LoRA weights into the off-the-shelf text-to-image model, DiffLoRA enables zero-shot personalization during inference, eliminating the need for post-processing optimization. Moreover, we introduce a novel identity-oriented LoRA weights construction pipeline to facilitate the training process of DiffLoRA. The dataset generated through this pipeline enables DiffLoRA to produce consistently high-quality LoRA weights. Notably, the distinctive properties of the diffusion model enhance the generation of superior weights by employing probabilistic modeling to capture intricate structural patterns and thoroughly explore the weight space. Comprehensive experimental results demonstrate that DiffLoRA outperforms existing personalization approaches across multiple benchmarks, achieving both time efficiency and maintaining identity fidelity throughout the personalization process.

Chengwei Sun, Jiwei Wei, Yujia Wu et al.

Large pre-trained models (LPMs) have demonstrated exceptional performance in diverse natural language processing and computer vision tasks. However, fully fine-tuning these models poses substantial memory challenges, particularly in resource-constrained environments. Parameter-efficient fine-tuning (PEFT) methods, such as LoRA, mitigate this issue by adjusting only a small subset of parameters. Nevertheless, these methods typically employ random initialization for low-rank matrices, which can lead to inefficiencies in gradient descent and diminished generalizability due to suboptimal starting points. To address these limitations, we propose SVFit, a novel PEFT approach that leverages singular value decomposition (SVD) to initialize low-rank matrices using critical singular values as trainable parameters. Specifically, SVFit performs SVD on the pre-trained weight matrix to obtain the best rank-r approximation matrix, emphasizing the most critical singular values that capture over 99% of the matrix's information. These top-r singular values are then used as trainable parameters to scale the fundamental subspaces of the matrix, facilitating rapid domain adaptation. Extensive experiments across various pre-trained models in natural language understanding, text-to-image generation, and image classification tasks reveal that SVFit outperforms LoRA while requiring 16 times fewer trainable parameters.

GLM-V Team, Wenyi Hong, Wenmeng Yu et al.

We present GLM-4.1V-Thinking and GLM-4.5V, a family of vision-language models (VLMs) designed to advance general-purpose multimodal understanding and reasoning. In this report, we share our key findings in the development of the reasoning-centric training framework. We first develop a capable vision foundation model with significant potential through large-scale pre-training, which arguably sets the upper bound for the final performance. We then propose Reinforcement Learning with Curriculum Sampling (RLCS) to unlock the full potential of the model, leading to comprehensive capability enhancement across a diverse range of tasks, including STEM problem solving, video understanding, content recognition, coding, grounding, GUI-based agents, and long document interpretation. In a comprehensive evaluation across 42 public benchmarks, GLM-4.5V achieves state-of-the-art performance on nearly all tasks among open-source models of similar size, and demonstrates competitive or even superior results compared to closed-source models such as Gemini-2.5-Flash on challenging tasks including Coding and GUI Agents. Meanwhile, the smaller GLM-4.1V-9B-Thinking remains highly competitive-achieving superior results to the much larger Qwen2.5-VL-72B on 29 benchmarks. We open-source both GLM-4.1V-9B-Thinking and GLM-4.5V. Code, models and more information are released at https://github.com/zai-org/GLM-V.

Yixiao Yuan, Yangchen Huang, Yu Ma et al.

Neural language representation models such as GPT, pre-trained on large-scale corpora, can effectively capture rich semantic patterns from plain text and be fine-tuned to consistently improve natural language generation performance. However, existing pre-trained language models used to generate lyrics rarely consider rhyme information, which is crucial in lyrics. Using a pre-trained model directly results in poor performance. To enhance the rhyming quality of generated lyrics, we incorporate integrated rhyme information into our model, thereby improving lyric generation performance.

Junlong Jia, Ying Hu, Xi Weng et al.

We present TinyLLaVA Factory, an open-source modular codebase for small-scale large multimodal models (LMMs) with a focus on simplicity of code implementations, extensibility of new features, and reproducibility of training results. Following the design philosophy of the factory pattern in software engineering, TinyLLaVA Factory modularizes the entire system into interchangeable components, with each component integrating a suite of cutting-edge models and methods, meanwhile leaving room for extensions to more features. In addition to allowing users to customize their own LMMs, TinyLLaVA Factory provides popular training recipes to let users pretrain and finetune their models with less coding effort. Empirical experiments validate the effectiveness of our codebase. The goal of TinyLLaVA Factory is to assist researchers and practitioners in exploring the wide landscape of designing and training small-scale LMMs with affordable computational resources.

Yiming Shi, Xun Zhu, Kaiwen Wang et al.

3D medical image analysis is essential for modern healthcare, yet traditional task-specific models are inadequate due to limited generalizability across diverse clinical scenarios. Multimodal large language models (MLLMs) offer a promising solution to these challenges. However, existing MLLMs have limitations in fully leveraging the rich, hierarchical information embedded in 3D medical images. Inspired by clinical practice, where radiologists focus on both 3D spatial structure and 2D planar content, we propose Med-2E3, a 3D medical MLLM that integrates a dual 3D-2D encoder architecture. To aggregate 2D features effectively, we design a Text-Guided Inter-Slice (TG-IS) scoring module, which scores the attention of each 2D slice based on slice contents and task instructions. To the best of our knowledge, Med-2E3 is the first MLLM to integrate both 3D and 2D features for 3D medical image analysis. Experiments on large-scale, open-source 3D medical multimodal datasets demonstrate that TG-IS exhibits task-specific attention distribution and significantly outperforms current state-of-the-art models. The code is available at: https://github.com/MSIIP/Med-2E3

Yiming Shi, Jiwei Wei, Yujia Wu et al.

The rapid growth of model scale has necessitated substantial computational resources for fine-tuning. Existing approach such as Low-Rank Adaptation (LoRA) has sought to address the problem of handling the large updated parameters in full fine-tuning. However, LoRA utilize random initialization and optimization of low-rank matrices to approximate updated weights, which can result in suboptimal convergence and an accuracy gap compared to full fine-tuning. To address these issues, we propose LoLDU, a Parameter-Efficient Fine-Tuning (PEFT) approach that significantly reduces trainable parameters by 2600 times compared to regular PEFT methods while maintaining comparable performance. LoLDU leverages Lower-Diag-Upper Decomposition (LDU) to initialize low-rank matrices for faster convergence and orthogonality. We focus on optimizing the diagonal matrix for scaling transformations. To the best of our knowledge, LoLDU has the fewest parameters among all PEFT approaches. We conducted extensive experiments across 4 instruction-following datasets, 6 natural language understanding (NLU) datasets, 8 image classification datasets, and image generation datasets with multiple model types (LLaMA2, RoBERTa, ViT, and Stable Diffusion), providing a comprehensive and detailed analysis. Our open-source code can be accessed at \href{https://github.com/SKDDJ/LoLDU}{https://github.com/SKDDJ/LoLDU}.

Hongjian Yu, Yiming Shi, Zherui Zhou et al.

We introduce a FLORES+ dataset as an evaluation benchmark for modern Wu Chinese machine translation models and showcase its compatibility with existing Wu data. Wu Chinese is mutually unintelligible with other Sinitic languages such as Mandarin and Yue (Cantonese), but uses a set of Hanzi (Chinese characters) that profoundly overlaps with others. The population of Wu speakers is the second largest among languages in China, but the language has been suffering from significant drop in usage especially among the younger generations. We identify Wu Chinese as a textually low-resource language and address challenges for its machine translation models. Our contributions include: (1) an open-source, manually translated dataset, (2) full documentations on the process of dataset creation and validation experiments, (3) preliminary tools for Wu Chinese normalization and segmentation, and (4) benefits and limitations of our dataset, as well as implications to other low-resource languages.

Yiming Shi, Shaoshuai Yang, Xun Zhu et al.

Medical image analysis is essential in modern healthcare. Deep learning has redirected research focus toward complex medical multimodal tasks, including report generation and visual question answering. Traditional task-specific models often fall short in handling these challenges. Large vision-language models (LVLMs) offer new solutions for solving such tasks. In this study, we build on the popular LLaVA framework to systematically explore model architectures and training strategies for both 2D and 3D medical LVLMs. We present extensive empirical findings and practical guidance. To support reproducibility and future research, we release a modular codebase, MedM-VL, and two pre-trained models: MedM-VL-2D for 2D medical image analysis and MedM-VL-CT-Chest for 3D CT-based applications. The code is available at: https://github.com/MSIIP/MedM-VL

Xun Zhu, Fanbin Mo, Xi Chen et al.

The rise of multimodal large language models (MLLMs) has sparked an unprecedented wave of applications in the field of medical imaging analysis. However, as one of the earliest and most fundamental tasks integrated into this paradigm, medical image classification reveals a sobering reality: state-of-the-art medical MLLMs consistently underperform compared to traditional deep learning models, despite their overwhelming advantages in pre-training data and model parameters. This paradox prompts a critical rethinking: where exactly does the performance degradation originate? In this paper, we conduct extensive experiments on 14 open-source medical MLLMs across three representative image classification datasets. Moving beyond superficial performance benchmarking, we employ feature probing to track the information flow of visual features module-by-module and layer-by-layer throughout the entire MLLM pipeline, enabling explicit visualization of where and how classification signals are distorted, diluted, or overridden. As the first attempt to dissect classification performance degradation in medical MLLMs, our findings reveal four failure modes: 1) quality limitation in visual representation, 2) fidelity loss in connector projection, 3) comprehension deficit in LLM reasoning, and 4) misalignment of semantic mapping. Meanwhile, we introduce quantitative scores that characterize the healthiness of feature evolution, enabling principled comparisons across diverse MLLMs and datasets. Furthermore, we provide insightful discussions centered on the critical barriers that prevent current medical MLLMs from fulfilling their promised clinical potential. We hope that our work provokes rethinking within the community-highlighting that the road from high expectations to clinically deployable MLLMs remains long and winding.

Kaiwen Wang, Kaili Zheng, Yiming Shi et al.

Multi-person human mesh recovery from a single image is a challenging task, hindered by the scarcity of in-the-wild training data. Prevailing in-the-wild human mesh pseudo-ground-truth (pGT) generation pipelines are single-person-centric, where each human is processed individually without joint optimization. This oversight leads to a lack of scene-level consistency, producing individuals with conflicting depths and scales within the same image. To address this, we introduce Depth-conditioned Translation Optimization (DTO), a novel optimization-based method that jointly refines the camera-space translations of all individuals in a crowd. By leveraging anthropometric priors on human height and depth cues from a monocular depth estimator, DTO solves for a scene-consistent placement of all subjects within a principled Maximum a posteriori (MAP) framework. Applying DTO to the 4D-Humans dataset, we construct DTO-Humans, a new large-scale pGT dataset of 0.56M high-quality, scene-consistent multi-person images, featuring dense crowds with an average of 4.8 persons per image. Furthermore, we propose Metric-Aware HMR, an end-to-end network that directly estimates human mesh and camera parameters in metric scale. This is enabled by a camera branch and a relative metric loss that enforces plausible relative scales. Extensive experiments demonstrate that our method achieves state-of-the-art performance on relative depth reasoning and human mesh recovery. Code is available at: https://github.com/gouba2333/MA-HMR.

Xun Zhu, Fanbin Mo, Zheng Zhang et al.

The emergence of medical generalist foundation models has revolutionized conventional task-specific model development paradigms, aiming to better handle multiple tasks through joint training on large-scale medical datasets. However, recent advances prioritize simple data scaling or architectural component enhancement, while neglecting to re-examine multi-task learning from a data-centric perspective. Critically, simply aggregating existing data resources leads to decentralized image-task alignment, which fails to cultivate comprehensive image understanding or align with clinical needs for multi-dimensional image interpretation. In this paper, we introduce the image-centric multi-annotation X-ray dataset (IMAX), the first attempt to enhance the multi-task learning capabilities of medical multi-modal large language models (MLLMs) from the data construction level. To be specific, IMAX is featured from the following attributes: 1) High-quality data curation. A comprehensive collection of more than 354K entries applicable to seven different medical tasks. 2) Image-centric dense annotation. Each X-ray image is associated with an average of 4.10 tasks and 7.46 training entries, ensuring multi-task representation richness per image. Compared to the general decentralized multi-annotation X-ray dataset (DMAX), IMAX consistently demonstrates significant multi-task average performance gains ranging from 3.20% to 21.05% across seven open-source state-of-the-art medical MLLMs. Moreover, we investigate differences in statistical patterns exhibited by IMAX and DMAX training processes, exploring potential correlations between optimization dynamics and multi-task performance. Finally, leveraging the core concept of IMAX data construction, we propose an optimized DMAX-based training strategy to alleviate the dilemma of obtaining high-quality IMAX data in practical scenarios.

Yao Pu, Yiming Shi, Zhenxi Zhang et al.

Magnetic resonance imaging (MRI) is essential for nasopharyngeal carcinoma (NPC) radiotherapy (RT), but practical constraints, such as patient discomfort, long scan times, and high costs often lead to incomplete modalities in clinical practice, compromising RT planning accuracy. Traditional MRI synthesis methods are modality-specific, limited in anatomical adaptability, and lack clinical interpretability-failing to meet NPC's RT needs. Here, we developed a unified foundation model integrating contrastive visual representation learning and vision-language alignment (VLA) to enable any-to-all MRI synthesis. The model uses a contrastive encoder for modality-invariant representations and a CLIP-based text-informed decoder for semantically consistent synthesis, supporting any-to-all MRI synthesis via one unified foundation model. Trained on 40,825 images from 13 institutions, it achieves consistently high performance (average SSIM 0.90, PSNR 27) across 26 internal/external validation sites (15,748 images), with superior synthesis fidelity and robustness to noise and domain shifts. Meanwhile, its unified representation enhances downstream RT-relevant tasks (e.g., segmentation). This work advances digital medicine solutions for NPC care by leveraging foundation models to bridge technical synthesis and clinical utility.

Xun Zhu, Zheng Zhang, Xi Chen et al.

With the rapid advancements in multi-modal large language models (MLLMs), connectors play a pivotal role in bridging diverse modalities and enhancing model performance. However, the design and evolution of connectors have not been comprehensively analyzed, leaving gaps in understanding how these components function and hindering the development of more powerful connectors. In this survey, we systematically review the current progress of connectors in MLLMs and present a structured taxonomy that categorizes connectors into atomic operations (mapping, compression, mixture of experts) and holistic designs (multi-layer, multi-encoder, multi-modal scenarios), highlighting their technical contributions and advancements. Furthermore, we discuss several promising research frontiers and challenges, including high-resolution input, dynamic compression, guide information selection, combination strategy, and interpretability. This survey is intended to serve as a foundational reference and a clear roadmap for researchers, providing valuable insights into the design and optimization of next-generation connectors to enhance the performance and adaptability of MLLMs.

Kaiwen Wang, Kaili Zheng, Rongrong Deng et al.

Recent multimodal large language models (MLLMs) have shown strong capabilities in general video understanding, driving growing interest in automatic sports commentary generation. However, existing benchmarks for this task focus exclusively on team sports such as soccer and basketball, leaving combat sports entirely unexplored. Notably, combat sports present distinct challenges: critical actions unfold within milliseconds with visually subtle yet semantically decisive differences, and professional commentary contains a substantially higher proportion of tactical analysis compared to team sports. In this paper, we present BoxComm, a large-scale dataset comprising 445 World Boxing Championship match videos with over 52K commentary sentences from professional broadcasts. We propose a structured commentary taxonomy that categorizes each sentence into play-by-play, tactical, or contextual, providing the first category-level annotation for sports commentary benchmarks. Building on this taxonomy, we introduce two novel and complementary evaluations tailored to sports commentary generation: (1) category-conditioned generation, which evaluates whether models can produce accurate commentary of a specified type given video context; and (2) commentary rhythm assessment, which measures whether freely generated commentary exhibits appropriate temporal pacing and type distribution over continuous video segments, capturing a dimension of commentary competence that prior benchmarks have not addressed. Experiments on multiple state-of-the-art MLLMs reveal that current models struggle on both evaluations. We further propose EIC-Gen, an improved baseline incorporating detected punch events to supply structured action cues, yielding consistent gains and highlighting the importance of perceiving fleeting and subtle events for combat sports commentary.

Duo Su, Huyu Wu, Huanran Chen et al.

Dataset distillation aims to synthesize compact yet informative datasets from large ones. A significant challenge in this field is achieving a trifecta of diversity, generalization, and representativeness in a single distilled dataset. Although recent generative dataset distillation methods adopt powerful diffusion models as their foundation models, the inherent representativeness prior in diffusion models is overlooked. Consequently, these approaches often necessitate the integration of external constraints to enhance data quality. To address this, we propose Diffusion As Priors (DAP), which formalizes representativeness by quantifying the similarity between synthetic and real data in feature space using a Mercer kernel. We then introduce this prior as guidance to steer the reverse diffusion process, enhancing the representativeness of distilled samples without any retraining. Extensive experiments on large-scale datasets, such as ImageNet-1K and its subsets, demonstrate that DAP outperforms state-of-the-art methods in generating high-fidelity datasets while achieving superior cross-architecture generalization. Our work not only establishes a theoretical connection between diffusion priors and the objectives of dataset distillation but also provides a practical, training-free framework for improving the quality of the distilled dataset.