Document understanding aims to perform question answering and information extraction over document images, where the visual content is highly information-dense and most queries rely on only a few relevant layout regions. However, existing methods either adopt a one-pass strategy that implicitly assumes all layouts are equally important, or focus excessively on small regions at the cost of losing critical layout information. To address these limitations, we introduce Doc-CoB (Chain-of-Boxes), a simple-yet-effective framework that integrates coarse-to-fine layout-aware visual reasoning into multimodal large language models. Instead of directly zooming into small regions, Doc-CoB progressively focuses on query-relevant layouts while preserving global document information. Specifically, it first selects key layout boxes and then focuses on them for further understanding with visual prompting. To support this paradigm, we introduce two reasoning tasks for box recognition and box reasoning, with an automatic pipeline that constructs 249k training samples with intermediate visual supervision. Experiments on seven benchmarks with four popular models show that Doc-CoB significantly improves performance, demonstrating its effectiveness and wide applicability. The code and the data are available at https://github.com/Doc-CoB/Doc-CoB.

Autonomous Graphical User Interface (GUI) agents powered by Multimodal Large Language Models (MLLMs) enable digital automation on end-user devices. While scaling both parameters and data has yielded substantial gains, advanced methods still suffer from prohibitive deployment costs on resource-constrained devices. When facing complex in-the-wild scenarios, lightweight GUI agents are bottlenecked by limited capacity and poor task scalability under end-to-end episodic learning, impeding multi-agent systems (MAS) adaptation, while training multiple skill-specific experts remains costly. Can we strike an effective trade-off in this cost–scalability dilemma, enabling lightweight MLLMs to participate in realistic GUI workflows? To address these challenges, we propose LAMO framework, which endows a lightweight MLLM with GUI-specific knowledge and task scalability, allowing multi-role orchestration to expand their capability boundary for GUI automation. LAMO combines role-oriented data synthesis with a two-stage training recipe: (i) supervised fine-tuning with Perplexity-Weighted Cross-Entropy optimization for knowledge distillation and visual perception enhancement, and (ii) reinforcement learning for role-oriented cooperative exploration. Via LAMO, we develop a task-scalable native GUI agent LAMO-3B supporting monolithic execution and MAS-style orchestration. When paired with advanced planners, as a plug-and-play policy executor, LAMO-3B can continuously benefit from planner advances, enabling a higher performance ceiling. Extensive static and online evaluations validate the effectiveness of our designs.

Knowledge-based Visual Question Answering (KB-VQA) requires models to answer questions by integrating visual information with external knowledge. However, retrieved knowledge is often noisy, partially irrelevant, or misaligned with the visual content, while internal model knowledge is difficult to control and interpret. Naïve aggregation of these sources limits reasoning effectiveness and reduces answer accuracy. To address this, we propose MaS-VQA, a selection-driven framework that tightly couples explicit knowledge filtering with implicit knowledge reasoning. MaS-VQA first retrieves candidate passages and applies a Mask-and-Select mechanism to jointly prune irrelevant image regions and weakly relevant knowledge fragments, producing compact, high-signal multimodal knowledge . This filtered knowledge then guides the activation of internal knowledge in a constrained semantic space, enabling complementary co-modeling of explicit and implicit knowledge for robust answer prediction. Experiments on Encyclopedic-VQA and InfoSeek demonstrate consistent performance gains across multiple MLLM backbones, and ablations verify that the selection mechanism effectively reduces noise and enhances knowledge utilization.

Audio Description (AD) aims to generate narrations of information that is not accessible through unimodal hearing in movies to aid the visually impaired in following film narratives. Current solutions rely heavily on manual work, resulting in high costs and limited scalability. While automatic methods have been introduced, they often yield descriptions that are sparse and omit key details. ddressing these challenges, we propose a novel automated pipeline, the Multi-modal Movie Audio Description (MMAD). MMAD harnesses the capabilities of three key modules as well as the power of Llama2 to augment the depth and breadth of the generated descriptions. Specifically, first, we propose an Audio-aware Feature Enhancing Module to provide the model with multi-modal perception capabilities, enriching the background descriptions with a more comprehensive understanding of the environmental features. Second, we propose an Actor-tracking-aware Story Linking Module to aid in the generation of contextual and character-centric descriptions, thereby enhancing the richness of character depictions. Third, we incorporate a Subtitled Movie Clip Contextual Alignment Module, supplying semantic information about various time periods throughout the movie, which facilitates the consideration of the full movie narrative context when describing silent segments, thereby enhancing the richness of the descriptions. Experiments on widely used datasets convincingly demonstrates that MMAD significantly surpasses existing strong baselines in performance, establishing a new state-of-the-art in the field. Our code will be released at https://github.com/Daria8976/MMAD.