Jihao Wu

Ning Wang, Jiangrong Xie, Hang Luo et al.

Recent years have witnessed the rapid progress of image captioning. However, the demands for large memory storage and heavy computational burden prevent these captioning models from being deployed on mobile devices. The main obstacles lie in the heavyweight visual feature extractors (i.e., object detectors) and complicated cross-modal fusion networks. To this end, we propose LightCap, a lightweight image captioner for resource-limited devices. The core design is built on the recent CLIP model for efficient image captioning. To be specific, on the one hand, we leverage the CLIP model to extract the compact grid features without relying on the time-consuming object detectors. On the other hand, we transfer the image-text retrieval design of CLIP to image captioning scenarios by devising a novel visual concept extractor and a cross-modal modulator. We further optimize the cross-modal fusion model and parallel prediction heads via sequential and ensemble distillations. With the carefully designed architecture, our model merely contains 40M parameters, saving the model size by more than 75% and the FLOPs by more than 98% in comparison with the current state-of-the-art methods. In spite of the low capacity, our model still exhibits state-of-the-art performance on prevalent datasets, e.g., 136.6 CIDEr on COCO Karpathy test split. Testing on the smartphone with only a single CPU, the proposed LightCap exhibits a fast inference speed of 188ms per image, which is ready for practical applications.

Ya-Qi Yu, Hao Wang, Fangyu Hong et al.

While Reinforcement Learning with Verifiable Rewards (RLVR) is effective for deterministically checkable tasks, many vision-language tasks are partially verifiable, demanding multi-criteria supervision (e.g., perceptual details, reasoning steps, and constraints). Rubrics provide a natural interface for this fine-grained supervision, but their effectiveness depends on the execution accuracy during online RL. We propose Reinforcement Learning with Robust Rubric Rewards ($\text{RLR}^3$), extending RLVR from task-level verification to criterion-level verification. $\text{RLR}^3$ routes instance-specific rubrics through two execution paths: an LLM-as-an-extractor paired with a deterministic verifier, or an LLM-as-a-Judge for non-verifiable criteria. To ensure faithful scoring, $\text{RLR}^3$ introduce a minimal exposure strategy that masks ground truths from extractors and images from judges. Furthermore, $\text{RLR}^3$ employs hierarchical aggregation to prioritize essential criteria over additional criteria, and mitigates score saturation within rollout groups. Evaluated on Qwen3-VL-30B-A3B across 15 benchmarks, $\text{RLR}^3$ consistently outperforms RLVR, yielding a 4.7-point improvement over the base model and exceeding the official instruct-to-thinking model gap. Controlled audits confirm our deterministic verification and minimal exposure significantly reduce exploitable false positives.

Ning Wang, Jiahao Xie, Jihao Wu et al.

Despite the remarkable progress of image captioning, existing captioners typically lack the controllable capability to generate desired image captions, e.g., describing the image in a rough or detailed manner, in a factual or emotional view, etc. In this paper, we show that a unified model is qualified to perform well in diverse domains and freely switch among multiple styles. Such a controllable capability is achieved by embedding the prompt learning into the image captioning framework. To be specific, we design a set of prompts to fine-tune the pre-trained image captioner. These prompts allow the model to absorb stylized data from different domains for joint training, without performance degradation in each domain. Furthermore, we optimize the prompts with learnable vectors in the continuous word embedding space, avoiding the heuristic prompt engineering and meanwhile exhibiting superior performance. In the inference stage, our model is able to generate desired stylized captions by choosing the corresponding prompts. Extensive experiments verify the controllable capability of the proposed method. Notably, we achieve outstanding performance on two diverse image captioning benchmarks including COCO Karpathy split and TextCaps using a unified model.

Hao Yan, Yuliang Liu, Xingchen Liu et al.

Existing Multimodal Large Language Models (MLLMs) suffer from significant performance degradation on the long document understanding task as document length increases. This stems from two fundamental challenges: 1) a low Signal-to-Noise Ratio (SNR), with crucial evidence buried in irrelevant pages; and 2) supervision scarcity, as datasets offering only final short answers provide a weak learning signal. In this paper, we address these challenges by proposing a paradigm that requires the model to execute a structured Analysis, Localization and Reasoning workflow. To instill this capability, we design a two-stage training framework: we first perform Supervised Fine-Tuning on high-quality data generated via an efficient knowledge distillation strategy. Subsequently, we employ an Evidence-aware Group Relative Policy Optimization which jointly optimizes for both evidence localization and answer accuracy. Additionally, we introduce a Evidence-Guided Resolution Allocation strategy to mitigate memory constraints of training on multi-pages documents. Extensive experiments demonstrate that DocSeeker achieves superior performance on both in-domain and out-of-domain tasks. We show it robustly generalizes from short-page training to ultra-long documents and is naturally synergistic with visual Retrieval-Augmented Generation systems, serving as a solid foundation for their implementation.

Ya-Qi Yu, Fangyu Hong, Xiangyang Qu et al.

The effectiveness of Direct Preference Optimization (DPO) depends on preference data that reflect the quality differences that matter in multimodal tasks. Existing pipelines often rely on off-policy perturbations or coarse outcome-based signals, which are not well suited to fine-grained visual reasoning. We propose rDPO, a preference optimization framework based on instance-specific rubrics. For each image-instruction pair, we create a checklist-style rubric of essential and additional criteria to score responses from any possible policies. The instruction-rubric pool is built offline and reused during the construction of on-policy data. On public reward modeling benchmarks, rubric-based prompting massively improves a 30B-A3B judge and brings it close to GPT-5.4. On public downstream benchmarks, rubric-based filtering raises the macro average to 82.69, whereas outcome-based filtering drops it to 75.82 from 81.14. When evaluating scalability on a comprehensive benchmark, rDPO achieves 61.01, markedly outperforming the style-constrained baseline (52.36) and surpassing the 59.48 base model. Together, these results show that visual preference optimization benefits from combining on-policy data construction with instance-specific criterion-level feedback.

Chaowei Liu, Jichun Li, Yihua Teng et al.

For capturing colored document images, e.g. posters and magazines, it is common that multiple degradations such as shadows, wrinkles, etc., are simultaneously introduced due to external factors. Restoring multi-degraded colored document images is a great challenge, yet overlooked, as most existing algorithms focus on enhancing color-ignored document images via binarization. Thus, we propose DocStormer, a novel algorithm designed to restore multi-degraded colored documents to their potential pristine PDF. The contributions are: firstly, we propose a "Perceive-then-Restore" paradigm with a reinforced transformer block, which more effectively encodes and utilizes the distribution of degradations. Secondly, we are the first to utilize GAN and pristine PDF magazine images to narrow the distribution gap between the enhanced results and PDF images, in pursuit of less degradation and better visual quality. Thirdly, we propose a non-parametric strategy, PFILI, which enables a smaller training scale and larger testing resolutions with acceptable detail trade-off, while saving memory and inference time. Fourthly, we are the first to propose a novel Multi-Degraded Colored Document image Enhancing dataset, named MD-CDE, for both training and evaluation. Experimental results show that the DocStormer exhibits superior performance, capable of revitalizing multi-degraded colored documents into their potential pristine digital versions, which fills the current academic gap from the perspective of method, data, and task.

Hao Yan, Xingchen Liu, Hao Wang et al.

Recent strides in multimodal large language models (MLLMs) have significantly advanced their performance in many reasoning tasks. However, Abstract Visual Reasoning (AVR) remains a critical challenge, primarily due to limitations in perceiving abstract graphics. To tackle this issue, we investigate the bottlenecks in current MLLMs and synthesize training data to improve their abstract visual perception. First, we propose VisuRiddles, a benchmark for AVR, featuring tasks meticulously constructed to assess models' reasoning capacities across five core dimensions and two high-level reasoning categories. Second, we introduce the Perceptual Riddle Synthesizer (PRS), an automated framework for generating riddles with fine-grained perceptual descriptions. PRS not only generates valuable training data for abstract graphics but also provides fine-grained perceptual description, crucially allowing for supervision over intermediate reasoning stages and thereby improving both training efficacy and model interpretability. Our extensive experimental results on VisuRiddles empirically validate that fine-grained visual perception is the principal bottleneck and our synthesis framework markedly enhances the performance of contemporary MLLMs on these challenging tasks. Our code and dataset will be released at https://github.com/yh-hust/VisuRiddles

Jiwen Zhang, Jihao Wu, Yihua Teng et al.

Large language model (LLM) leads to a surge of autonomous GUI agents for smartphone, which completes a task triggered by natural language through predicting a sequence of actions of API. Even though the task highly relies on past actions and visual observations, existing studies typically consider little semantic information carried out by intermediate screenshots and screen operations. To address this, this work presents Chain-of-Action-Thought (dubbed CoAT), which takes the description of the previous actions, the current screen, and more importantly the action thinking of what actions should be performed and the outcomes led by the chosen action. We demonstrate that, in a zero-shot setting upon three off-the-shelf LMMs, CoAT significantly improves the action prediction compared to previous proposed context modeling. To further facilitate the research in this line, we construct a dataset Android-In-The-Zoo (AitZ), which contains 18,643 screen-action pairs together with chain-of-action-thought annotations. Experiments show that fine-tuning a 1B model (i.e. AUTO-UI-base) on our AitZ dataset achieves on-par performance with CogAgent-Chat-18B.

Xijia Tao, Yihua Teng, Xinxing Su et al.

Existing multimodal browsing benchmarks often fail to require genuine multimodal reasoning, as many tasks can be solved with text-only heuristics without vision-in-the-loop verification. We introduce MMSearch-Plus, a 311-task benchmark that enforces multimodal understanding by requiring extraction and propagation of fine-grained visual cues through iterative image-text retrieval and cross-validation under retrieval noise. Our curation procedure seeds questions whose answers require extrapolating from spatial cues and temporal traces to out-of-image facts such as events, dates, and venues. Beyond the dataset, we provide a model-agnostic agent framework with standard browsing tools and a set-of-mark (SoM) module, which lets the agent place marks, crop subregions, and launch targeted image/text searches. SoM enables provenance-aware zoom-and-retrieve and improves robustness in multi-step reasoning. We evaluated closed- and open-source MLLMs in this framework. The strongest system achieves an end-to-end accuracy of 36.0%, and integrating SoM produces consistent gains in multiple settings, with improvements up to +3.9 points. From failure analysis, we observe recurring errors in locating relevant webpages and distinguishing between visually similar events. These results underscore the challenges of real-world multimodal search and establish MMSearch-Plus as a rigorous benchmark for advancing agentic MLLMs.

Xinmiao Yu, Xiaocheng Feng, Yun Li et al.

Recent Large Vision-Language Models (LVLMs) have shown promising reasoning capabilities on text-rich images from charts, tables, and documents. However, the abundant text within such images may increase the model's sensitivity to language. This raises the need to evaluate LVLM performance on cross-lingual text-rich visual inputs, where the language in the image differs from the language of the instructions. To address this, we introduce XT-VQA (Cross-Lingual Text-Rich Visual Question Answering), a benchmark designed to assess how LVLMs handle language inconsistency between image text and questions. XT-VQA integrates five existing text-rich VQA datasets and a newly collected dataset, XPaperQA, covering diverse scenarios that require faithful recognition and comprehension of visual information despite language inconsistency. Our evaluation of prominent LVLMs on XT-VQA reveals a significant drop in performance for cross-lingual scenarios, even for models with multilingual capabilities. A mutual information analysis suggests that this performance gap stems from cross-lingual questions failing to adequately activate relevant visual information. To mitigate this issue, we propose MVCL-MI (Maximization of Vision-Language Cross-Lingual Mutual Information), where a visual-text cross-lingual alignment is built by maximizing mutual information between the model's outputs and visual information. This is achieved by distilling knowledge from monolingual to cross-lingual settings through KL divergence minimization, where monolingual output logits serve as a teacher. Experimental results on the XT-VQA demonstrate that MVCL-MI effectively reduces the visual-text cross-lingual performance disparity while preserving the inherent capabilities of LVLMs, shedding new light on the potential practice for improving LVLMs. Codes are available at: https://github.com/Stardust-y/XTVQA.git

Ya-Qi Yu, Minghui Liao, Jihao Wu et al.

Multimodal Large Language Models (MLLMs) have shown impressive results on various multimodal tasks. However, most existing MLLMs are not well suited for document-oriented tasks, which require fine-grained image perception and information compression. In this paper, we present TextHawk, a MLLM that is specifically designed for document-oriented tasks, while preserving the general capabilities of MLLMs. TextHawk is aimed to explore efficient fine-grained perception by designing four dedicated components. Firstly, a ReSampling and ReArrangement (ReSA) module is proposed to reduce the redundancy in the document texts and lower the computational cost of the MLLM. We explore encoding the positions of each local feature by presenting Scalable Positional Embeddings (SPEs), which can preserve the scalability of various image sizes. A Query Proposal Network (QPN) is then adopted to initialize the queries dynamically among different sub-images. To further enhance the fine-grained visual perceptual ability of the MLLM, we design a Multi-Level Cross-Attention (MLCA) mechanism that captures the hierarchical structure and semantic relations of document images. Furthermore, we create a new instruction-tuning dataset for document-oriented tasks by enriching the multimodal document data with Gemini Pro. We conduct extensive experiments on both general and document-oriented MLLM benchmarks, and show that TextHawk outperforms the state-of-the-art methods, demonstrating its effectiveness and superiority in fine-grained document perception and general abilities.

Tianyi Zhuang, Chuqiao Kuang, Xiaoguang Li et al.

We present DocPuzzle, a rigorously constructed benchmark for evaluating long-context reasoning capabilities in large language models (LLMs). This benchmark comprises 100 expert-level QA problems requiring multi-step reasoning over long real-world documents. To ensure the task quality and complexity, we implement a human-AI collaborative annotation-validation pipeline. DocPuzzle introduces an innovative evaluation framework that mitigates guessing bias through checklist-guided process analysis, establishing new standards for assessing reasoning capacities in LLMs. Our evaluation results show that: 1)Advanced slow-thinking reasoning models like o1-preview(69.7%) and DeepSeek-R1(66.3%) significantly outperform best general instruct models like Claude 3.5 Sonnet(57.7%); 2)Distilled reasoning models like DeepSeek-R1-Distill-Qwen-32B(41.3%) falls far behind the teacher model, suggesting challenges to maintain the generalization of reasoning capabilities relying solely on distillation.

Feilong Chen, Yijiang Liu, Yi Huang et al.

We propose MindVL, a multimodal large language model (MLLMs) trained on Ascend NPUs. The training of state-of-the-art MLLMs is often confined to a limited set of hardware platforms and relies heavily on massive, undisclosed data recipes, which hinders reproducibility and open research. To change the common perception that Ascend hardware is unsuitable for efficient full-stage MLLM training, we introduce MindSpeed-MLLM, a highly efficient training framework that supports stable and high-performance training of large-scale Dense and Mixture-of-Experts (MoE) models on Ascend hardware. Based on this, we provide a systematic and open description of the data production methods and mixing strategies for all training stages. Furthermore, we present MindVL, a data-efficient multimodal large language model trained end-to-end on Ascend NPUs. In addition, we find that averaging weights from checkpoints trained with different sequence lengths is particularly effective and yields further gains when combined with test-time resolution search. Our experiments demonstrate superior data efficiency: MindVL-8B matches the performance of Qwen2.5VL-7B using only 10\% of its training data, while our MoE model, MindVL-671B-A37B, matches Qwen2.5VL-72B using only 3\% of the Qwen2.5VL training data, and achieves comparable performance with other leading multimodal MoE models. Our work provides the community with a valuable hardware alternative, open data recipes, and effective performance-enhancing techniques.

Tao Hu, Honglong Zhang, Fan Zeng et al.

In the field of intracity freight transportation, changes in order volume are significantly influenced by temporal and spatial factors. When building subsidy and pricing strategies, predicting the causal effects of these strategies on order volume is crucial. In the process of calculating causal effects, confounding variables can have an impact. Traditional methods to control confounding variables handle data from a holistic perspective, which cannot ensure the precision of causal effects in specific temporal and spatial dimensions. However, temporal and spatial dimensions are extremely critical in the logistics field, and this limitation may directly affect the precision of subsidy and pricing strategies. To address these issues, this study proposes a technique based on flexible temporal-spatial grid partitioning. Furthermore, based on the flexible grid partitioning technique, we further propose a continuous entropy balancing method in the temporal-spatial domain, which named TS-EBCT (Temporal-Spatial Entropy Balancing for Causal Continue Treatments). The method proposed in this paper has been tested on two simulation datasets and two real datasets, all of which have achieved excellent performance. In fact, after applying the TS-EBCT method to the intracity freight transportation field, the prediction accuracy of the causal effect has been significantly improved. It brings good business benefits to the company's subsidy and pricing strategies.