While previous multimodal slow-thinking methods have demonstrated remarkable success in single-image understanding scenarios, their effectiveness becomes fundamentally constrained when extended to more complex multi-image comprehension tasks. This limitation stems from their predominant reliance on text-based intermediate reasoning processes. While for human, when engaging in sophisticated multi-image analysis, they typically perform two complementary cognitive operations: (1) continuous cross-image visual comparison through region-of-interest matching, and (2) dynamic memorization of critical visual concepts throughout the reasoning chain. Motivated by these observations, we propose the Complex Multi-Modal Chain-of-Thought (CMMCoT) framework, a multi-step reasoning framework that mimics human-like "slow thinking" for multi-image understanding. Our approach incorporates two key innovations: (1) The construction of interleaved multimodal multi-step reasoning chains, which utilize critical visual region tokens, extracted from intermediate reasoning steps, as supervisory signals. This mechanism not only facilitates comprehensive cross-modal understanding but also enhances model interpretability. (2) The introduction of a test-time memory augmentation module that expands the model’s reasoning capacity during inference while preserving parameter efficiency. Furthermore, to facilitate research in this direction, we have curated a novel multi-image slow-thinking dataset. Extensive experiments demonstrate the effectiveness of our model.

Auto-regressive models have made significant progress in the realm of text-to-image synthesis, yet devising an appropriate model architecture and training strategy to achieve a satisfactory level remains an important avenue of exploration. In this work, we introduce MARS, a novel framework for T2I generation that incorporates a specially designed Semantic Vision-Language Integration Expert (SemVIE). This innovative component integrates pre-trained LLMs by independently processing linguistic and visual information—freezing the textual component while fine-tuning the visual component. This methodology preserves the NLP capabilities of LLMs while imbuing them with exceptional visual understanding. Building upon the powerful base of the pre-trained Qwen-7B, MARS stands out with its bilingual generative capabilities corresponding to both English and Chinese language prompts and the capacity for joint image and text generation. The flexibility of this framework lends itself to migration towards any-to-any task adaptability. Furthermore, MARS employs a multi-stage training strategy that first establishes robust image-text alignment through complementary bidirectional tasks and subsequently concentrates on refining the T2I generation process, significantly augmenting text-image synchrony and the granularity of image details. Notably, MARS requires only 9% of the GPU days needed by SD1.5, yet it achieves remarkable results across a variety of benchmarks, illustrating the training efficiency and the potential for swift deployment in various applications.

The rapidly developing Large Vision Language Models (LVLMs) still face the hallucination phenomena where the generated responses do not align with the given contexts, significantly restricting the usages of LVLMs. Most previous work detects and mitigates hallucination at the coarse-grained level or requires expensive annotation (e.g., labeling by human experts or proprietary models). To address these issues, we propose detecting and mitigating hallucinations in LVLMs via fine-grained AI feedback. The basic idea is that we generate a small-size sentence-level hallucination annotation dataset by proprietary models, whereby we train a detection model which can perform sentence-level hallucination detection. Then, we propose a detect-then-rewrite pipeline to automatically construct preference dataset for hallucination mitigation training. Furthermore, we propose differentiating the severity of hallucinations, and introducing a Hallucination Severity-Aware Direct Preference Optimization (HSA-DPO) which prioritizes the mitigation of critical hallucination in LVLMs by incorporating the severity of hallucinations into preference learning. Extensive experiments on hallucination detection and mitigation benchmarks demonstrate that our method sets a new state-of-the-art in hallucination detection on MHaluBench, surpassing GPT-4V and Gemini, and reduces the hallucination rate by 36.1% on AMBER and 76.3% on Object HalBench compared to the base model.

Ziwei Huang, Wanggui He, Quanyu Long, Yandi Wang, Haoyuan Li, Zhelun Yu, Fangxun Shu, Weilong Dai, Hao Jiang, Fei Wu, Leilei Gan. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

While Parameter-Efficient Fine-Tuning (PEFT) methods like Low-Rank Adaptation (LoRA) effectively address resource constraints during fine-tuning, their performance often falls short, especially in multidimensional task scenarios. To address this issue, one straightforward solution is to introduce task-specific LoRA as domain experts, leveraging the modeling of multiple capabilities of experts and thus enhancing the general capability of multi-task learning.Although promising, these additional components often add complexity to the training and inference process, contravening the efficiency that PEFT is designed to deliver. Considering this, we introduce an innovative PEFT method, TeamLoRA, consisting of a collaboration and competition module for LoRA experts, thus achieving the right balance of effectiveness and efficiency:(i) For collaboration, we introduce a novel knowledge sharing and organization mechanism designed to optimize hierarchical learning while enhancing the efficiency of model training and inference.(ii) For competition, we propose leveraging a game-theoretic interaction mechanism for experts, encouraging experts to transfer their domain-specific knowledge while facing diverse downstream tasks, thus enhancing the performance.By doing so, TeamLoRA elegantly connects the experts as a “Team” with internal collaboration and competition, enabling a faster and more accurate PEFT paradigm. Meanwhile, we curate a Comprehensive Multi-Task Evaluation (CME) benchmark to thoroughly assess the capability of multi-task learning. Experiments conducted on our CME and other benchmarks indicate the effectiveness and efficiency of TeamLoRA. Our project is available at https://github.com/DCDmllm/TeamLoRA.

Finetuning-free personalized image generation can synthesize customized images without test-time finetuning, attracting wide research interest owing to its high efficiency. Current finetuning-free methods simply adopt a single training stage with a simple image reconstruction task, and they typically generate low-quality images inconsistent with the reference images during test-time. To mitigate this problem, inspired by the recent DPO (i.e., direct preference optimization) technique, this work proposes an additional training stage to improve the pre-trained personalized generation models. However, traditional DPO only determines the overall superiority or inferiority of two samples, which is not suitable for personalized image generation because the generated images are commonly inconsistent with the reference images only in some local image patches. To tackle this problem, this work proposes PatchDPO that estimates the quality of image patches within each generated image and accordingly trains the model. To this end, PatchDPO first leverages the pre-trained vision model with a proposed self-supervised training method to estimate the patch quality. Next, PatchDPO adopts a weighted training approach to train the model with the estimated patch quality, which rewards the image patches with high quality while penalizing the image patches with low quality. Experiment results demonstrate that PatchDPO significantly improves the performance of multiple pre-trained personalized generation models, and achieves state-of-the-art performance on both single-object and multi-object personalized image generation. Our code is available at https://github.com/hqhQAQ/PatchDPO.

Subject-driven image personalization has seen notable advancements, especially with the ReferenceNet paradigm, which excels in integrating reference image features for creative and commercial applications. However, current ReferenceNet implementations mainly function as latent-level feature extractors, limiting their potential. This restricts the delivery of suitable features to the denoising backbone across timesteps, resulting in suboptimal image consistency. In this paper, we revisit reference feature extraction and propose TFCustom, a framework that focuses on reference image features at different temporal and frequency levels. We introduce synchronized ReferenceNet to extract reference features while optimizing noise injection and denoising. We also propose a time-aware frequency refinement module that uses high- and low-frequency filters with time embeddings to adaptively select reference feature injection. Additionally, we introduce a reward-based loss to improve the similarity between reference objects and generated images. Experimental results show that TFCustom outperforms existing methods in single-object and multi-object reference generation, with significant improvements in textual details.

MLLM reasoning has drawn widespread research for its excellent problem-solving capability. Current reasoning methods fall into two types: PRM, which supervises the intermediate reasoning steps, and ORM, which supervises the final results. Recently, DeepSeek-R1 has challenged the traditional view that PRM outperforms ORM, which demonstrates strong generalization performance using an ORM method (i.e., GRPO). However, current MLLM's GRPO algorithms still struggle to handle challenging and complex multimodal reasoning tasks (e.g., mathematical reasoning). In this work, we reveal two problems that impede the performance of GRPO on the MLLM: Low data utilization and Text-bias. Low data utilization refers to that GRPO cannot acquire positive rewards to update the MLLM on difficult samples, and text-bias is a phenomenon that the MLLM bypasses image condition and solely relies on text condition for generation after GRPO training. To tackle these problems, this work proposes Hint-GRPO that improves data utilization by adaptively providing hints for samples of varying difficulty, and text-bias calibration that mitigates text-bias by calibrating the token prediction logits with image condition in test-time. Experiment results on three base MLLMs across eleven datasets demonstrate that our proposed methods advance the reasoning capability of original MLLM by a large margin, exhibiting superior performance to existing MLLM reasoning methods. Our code is available at https://github.com/hqhQAQ/Hint-GRPO.

We propose ReKV, a novel training-free approach that enables efficient streaming video question-answering (StreamingVQA), by seamlessly integrating with existing Video Large Language Models (Video-LLMs). Traditional VideoQA systems struggle with long videos, as they must process entire videos before responding to queries, and repeat this process for each new question. In contrast, our approach analyzes long videos in a streaming manner, allowing for prompt responses as soon as user queries are received. Building on a common Video-LLM, we first incorporate a sliding-window attention mechanism, ensuring that input frames attend to a limited number of preceding frames, thereby reducing computational overhead. To prevent information loss, we store processed video key-value caches (KV-Caches) in RAM and disk, reloading them into GPU memory as needed. Additionally, we introduce a retrieval method that leverages an external retriever or the parameters within Video-LLMs to retrieve only query-relevant KV-Caches, ensuring both efficiency and accuracy in question answering. ReKV enables the separation of video analyzing and question-answering across different processes and GPUs, significantly enhancing the efficiency of StreamingVQA. Through comprehensive experimentation, we validate the efficacy and practicality of our approach, which significantly boosts efficiency and enhances applicability over existing VideoQA models.

We introduce LLaVA-MoD, a novel framework designed to enable the efficient training of small-scale Multimodal Language Models ($s$-MLLM) distilling knowledge from large-scale MLLM ($l$-MLLM). Our approach tackles two fundamental challenges in MLLM distillation. First, we optimize the network structure of $s$-MLLM by integrating a sparse Mixture of Experts (MoE) architecture into the language model, striking a balance between computational efficiency and model expressiveness. Second, we propose a progressive knowledge transfer strategy for comprehensive knowledge transfer. This strategy begins with mimic distillation, where we minimize the Kullback-Leibler (KL) divergence between output distributions to enable $s$-MLLM to emulate $s$-MLLM's understanding. Following this, we introduce preference distillation via Preference Optimization (PO), where the key lies in treating $l$-MLLM as the reference model. During this phase, the $s$-MLLM's ability to discriminate between superior and inferior examples is significantly enhanced beyond $l$-MLLM, leading to a better $s$-MLLM that surpasses $l$-MLLM, particularly in hallucination benchmarks. Extensive experiments demonstrate that LLaVA-MoD surpasses existing works across various benchmarks while maintaining a minimal activated parameters and low computational costs. Remarkably, LLaVA-MoD-2B surpasses Qwen-VL-Chat-7B with an average gain of 8.8\%, using merely $0.3\%$ of the training data and 23\% trainable parameters. The results underscore LLaVA-MoD's ability to effectively distill comprehensive knowledge from its teacher model, paving the way for developing efficient MLLMs.

Recent advancements in text-to-image generation models have dramatically enhanced the generation of photorealistic images from textual prompts, leading to an increased interest in personalized text-to-image applications, particularly in multi-subject scenarios. However, these advances are hindered by two main challenges: firstly, the need to accurately maintain the details of each referenced subject in accordance with the textual descriptions; and secondly, the difficulty in achieving a cohesive representation of multiple subjects in a single image without introducing inconsistencies. To address these concerns, our research introduces the MS-Diffusion framework for layout-guided zero-shot image personalization with multi-subjects. This innovative approach integrates grounding tokens with the feature resampler to maintain detail fidelity among subjects. With the layout guidance, MS-Diffusion further improves the cross-attention to adapt to the multi-subject inputs, ensuring that each subject condition acts on specific areas. The proposed multi-subject cross-attention orchestrates harmonious inter-subject compositions while preserving the control of texts. Comprehensive quantitative and qualitative experiments affirm that this method surpasses existing models in both image and text fidelity, promoting the development of personalized text-to-image generation.

We present **HealthGPT**, a powerful Medical Large Vision-Language Model (Med-LVLM) that integrates medical visual comprehension and generation capabilities within a unified autoregressive paradigm. Our bootstrapping philosophy is to progressively adapt heterogeneous comprehension and generation knowledge to pre-trained Large Language Models (LLMs). This is achieved through a novel heterogeneous low-rank adaptation **(H-LoRA)** technique, which is complemented by a tailored hierarchical visual perception **(HVP)** approach and a three-stage learning strategy **(TLS)**. To effectively learn the HealthGPT, we devise a comprehensive medical domain-specific comprehension and generation dataset called **VL-Health**. Experimental results demonstrate exceptional performance and scalability of HealthGPT in medical visual unified tasks. Our project can be accessed at https://github.com/DCDmllm/HealthGPT.

The pre-trained neural models have recently achieved impressive performance in understanding multimodal content. However, it is still very challenging to pre-train neural models for video and language understanding, especially for Chinese video-language data, due to the following reasons. Firstly, existing video-language pre-training algorithms mainly focus on the co-occurrence of words and video frames, but ignore other valuable semantic and structure information of video-language content, e.g., sequential order and spatiotemporal relationships. Secondly, there exist conflicts between video sentence alignment and other proxy tasks. Thirdly, there is a lack of large-scale and high-quality Chinese video-language datasets (eg. including 10 million unique videos), which are the fundamental success conditions for pre-training techniques. In this work, we propose a novel video-language understanding framework named Victor, which stands for VIdeo-language understanding via Contrastive mulTimOdal pRe-training. Besides general proxy tasks such as masked language modeling, Victor constructs several novel proxy tasks under the contrastive learning paradigm, making the model be more robust and able to capture more complex multimodal semantic and structural relationships from different perspectives. Victor is trained on a large-scale Chinese video-language dataset, including over 10 million complete videos with corresponding high-quality textual descriptions. We apply the pre-trained Victor model to a series of downstream applications and demonstrate its superior performance, comparing against the state-of-the-art pre-training methods such as VideoBERT and UniVL.