Hongkai Chen

Hongkai Chen, Zixin Luo, Lei Zhou et al.

Generating robust and reliable correspondences across images is a fundamental task for a diversity of applications. To capture context at both global and local granularity, we propose ASpanFormer, a Transformer-based detector-free matcher that is built on hierarchical attention structure, adopting a novel attention operation which is capable of adjusting attention span in a self-adaptive manner. To achieve this goal, first, flow maps are regressed in each cross attention phase to locate the center of search region. Next, a sampling grid is generated around the center, whose size, instead of being empirically configured as fixed, is adaptively computed from a pixel uncertainty estimated along with the flow map. Finally, attention is computed across two images within derived regions, referred to as attention span. By these means, we are able to not only maintain long-range dependencies, but also enable fine-grained attention among pixels of high relevance that compensates essential locality and piece-wise smoothness in matching tasks. State-of-the-art accuracy on a wide range of evaluation benchmarks validates the strong matching capability of our method.

Siyang Jiang, Mu Yuan, Xiang Ji et al.

Multimodal human action recognition (HAR) leverages complementary sensors for activity classification. Beyond recognition, recent advances in large language models (LLMs) enable detailed descriptions and causal reasoning, motivating new tasks: human action understanding (HAU) and human action reasoning (HARn). However, most LLMs, especially large vision language models (LVLMs), struggle with non-RGB modalities such as depth, IMU, and mmWave due to the lack of large-scale data-caption resources. Existing HAR datasets mainly provide coarse data-label annotations, which are insufficient to capture fine-grained action dynamics needed for HAU and HARn. We consider two ground-truth pair types: (1) data label (discrete category) and (2) data caption (textual description). Naively generating captions from labels often lacks logical and spatiotemporal consistency. We introduce CUHK-X, a large-scale multimodal dataset and benchmark suite for HAR, HAU, and HARn. CUHK-X contains 58,445 samples covering 40 actions performed by 30 participants across two indoor environments. To improve caption consistency, we propose a prompt-based scene creation method that leverages LLMs to generate logically connected activity sequences, followed by human validation. CUHK-X includes three benchmarks with six evaluation tasks. Experiments report average accuracies of 76.52% (HAR), 40.76% (HAU), and 70.25% (HARn). CUHK-X aims to enable the community to apply and develop data-intensive learning methods for robust, multimodal human activity analysis. Project page and code: https://openaiotlab.github.io/CUHK-X/ and https://github.com/openaiotlab/CUHK-X.

Shiva Ketabi, Hongkai Chen, Haiwei Dong et al.

Various congestion control protocols have been designed to achieve high performance in different network environments. Modern online learning solutions that delegate the congestion control actions to a machine cannot properly converge in the stringent time scales of data centers. We leverage multiagent reinforcement learning to design a system for dynamic tuning of congestion control parameters at end-hosts in a data center. The system includes agents at the end-hosts to monitor and report the network and traffic states, and agents to run the reinforcement learning algorithm given the states. Based on the state of the environment, the system generates congestion control parameters that optimize network performance metrics such as throughput and latency. As a case study, we examine BBR, an example of a prominent recently-developed congestion control protocol. Our experiments demonstrate that the proposed system has the potential to mitigate the problems of static parameters.

Kushan Choksi, Hongkai Chen, Karan Joshi et al.

Recent research has demonstrated the capability of physiological signals to infer both user emotional and attention responses. This presents an opportunity for leveraging widely available physiological sensors in smartwatches, to detect real-time emotional cues in users, such as stress and excitement. In this paper, we introduce SensEmo, a smartwatch-based system designed for affective learning. SensEmo utilizes multiple physiological sensor data, including heart rate and galvanic skin response, to recognize a student's motivation and concentration levels during class. This recognition is facilitated by a personalized emotion recognition model that predicts emotional states based on degrees of valence and arousal. With real-time emotion and attention feedback from students, we design a Markov decision process-based algorithm to enhance student learning effectiveness and experience by by offering suggestions to the teacher regarding teaching content and pacing. We evaluate SensEmo with 22 participants in real-world classroom environments. Evaluation results show that SensEmo recognizes student emotion with an average of 88.9% accuracy. More importantly, SensEmo assists students to achieve better online learning outcomes, e.g., an average of 40.0% higher grades in quizzes, over the traditional learning without student emotional feedback.

Sifan Li, Yujun Cai, Hongkai Chen et al.

Generating structured, editable diagrams remains a significant challenge for contemporary large language models, despite their proficiency in general-purpose vector code generation. The primary difficulty lies in the structural fragility of the output; minor errors such as misaligned connector endpoints, text labels overlapping borders, or complex layouts drifting beyond the canvas boundaries render the resulting SVG files functionally unusable for professional applications. To address these issues, we introduce GeoSVG-RL, a specialized reinforcement learning framework designed for layout-constrained text-to-SVG generation. Unlike standard training objectives that rely solely on maximizing token-level likelihood, our approach optimizes the policy against explicit, executable geometric feedback. The model first produces a structured layout plan that serves as a geometric contract for the subsequent generation of the SVG code. This code is then rendered through a browser-backed verifier, enabling the calculation of fine-grained rewards across six critical dimensions: rendering validity, canvas fitting, precise anchor placement, text containment, graph consistency, and code cleanliness. We utilize Group Relative Policy Optimization (GRPO) to refine the model, sampling multiple candidates per prompt to facilitate updates based on relative quality. Starting from a supervised warm-start phase on synthetic data, GeoSVG-RL achieves substantial gains in structural reliability, particularly in arrow-anchor accuracy and text-in-box rates. Quantitative evaluations demonstrate that our method consistently outperforms current state-of-the-art systems in local geometric precision and the preservation of graph connectivity, providing a robust pathway toward automated yet reliable technical illustration.

Sirajum Munir, Hongkai Chen, Shiwei Fang et al.

With the rise of hailing services, people are increasingly relying on shared mobility (e.g., Uber, Lyft) drivers to pick up for transportation. However, such drivers and riders have difficulties finding each other in urban areas as GPS signals get blocked by skyscrapers, in crowded environments (e.g., in stadiums, airports, and bars), at night, and in bad weather. It wastes their time, creates a bad user experience, and causes more CO2 emissions due to idle driving. In this work, we explore the potential of Wi-Fi to help drivers to determine the street side of the riders. Our proposed system is called CarFi that uses Wi-Fi CSI from two antennas placed inside a moving vehicle and a data-driven technique to determine the street side of the rider. By collecting real-world data in realistic and challenging settings by blocking the signal with other people and other parked cars, we see that CarFi is 95.44% accurate in rider-side determination in both line of sight (LoS) and non-line of sight (nLoS) conditions, and can be run on an embedded GPU in real-time.

Lilin Xu, Bufang Yang, Siyang Jiang et al.

Procedural tasks with multiple ordered steps are ubiquitous in daily life. Recent advances in multimodal large language models (MLLMs) have enabled personal assistants that support daily activities. However, existing systems primarily provide reactive guidance triggered by user queries, or limited proactive assistance for isolated short-term events rather than long-horizon procedural tasks. In this work, we introduce Pro$^2$Assist, a step-aware proactive assistant that continuously tracks fine-grained task progress and reasons over the user's evolving state to provide timely assistance throughout tasks. Pro$^2$Assist leverages multimodal data from augmented reality (AR) glasses to achieve motion-based perception. It then extracts step-oriented procedural context from multi-scale temporal dynamics and task-specific expert knowledge. Based on both sensory input and procedural context, Pro$^2$Assist performs continuous reasoning to infer user needs and display timely assistance on AR glasses. We evaluate Pro$^2$Assist using a dataset curated from public sources and a real-world dataset collected on our testbed with AR glasses. Extensive evaluations show that Pro$^2$Assist outperforms the best-performing baselines by over 21% in procedural action understanding accuracy, and it achieves up to 2.29x the proactive timing accuracy of baselines. A user study with 20 participants further shows that 90% find Pro$^2$Assist useful, indicating its effectiveness for real-world procedural assistance.

Bufang Yang, Lilin Xu, Liekang Zeng et al.

Recent advances in Large Language Models (LLMs) have propelled intelligent agents from reactive responses to proactive support. While promising, existing proactive agents either rely exclusively on observations from enclosed environments (e.g., desktop UIs) with direct LLM inference or employ rule-based proactive notifications, leading to suboptimal user intent understanding and limited functionality for proactive service. In this paper, we introduce ContextAgent, the first context-aware proactive agent that incorporates extensive sensory contexts surrounding humans to enhance the proactivity of LLM agents. ContextAgent first extracts multi-dimensional contexts from massive sensory perceptions on wearables (e.g., video and audio) to understand user intentions. ContextAgent then leverages the sensory contexts and personas from historical data to predict the necessity for proactive services. When proactive assistance is needed, ContextAgent further automatically calls the necessary tools to assist users unobtrusively. To evaluate this new task, we curate ContextAgentBench, the first benchmark for evaluating context-aware proactive LLM agents, covering 1,000 samples across nine daily scenarios and twenty tools. Experiments on ContextAgentBench show that ContextAgent outperforms baselines by achieving up to 8.5% and 6.0% higher accuracy in proactive predictions and tool calling, respectively. We hope our research can inspire the development of more advanced, human-centric, proactive AI assistants. The code and dataset are publicly available at https://github.com/openaiotlab/ContextAgent.

Kaiwei Liu, Yuting He, Bufang Yang et al.

Biosignals collected from wearable devices are widely utilized in healthcare applications. Machine learning models used in these applications often rely on features extracted from biosignals due to their effectiveness, lower data dimensionality, and wide compatibility across various model architectures. However, existing feature extraction methods often lack task-specific contextual knowledge, struggle to identify optimal feature extraction settings in high-dimensional feature space, and are prone to code generation and automation errors. In this paper, we propose DeepFeature, the first LLM-empowered, context-aware feature generation framework for wearable biosignals. DeepFeature introduces a multi-source feature generation mechanism that integrates expert knowledge with task settings. It also employs an iterative feature refinement process that uses feature assessment-based feedback for feature re-selection. Additionally, DeepFeature utilizes a robust multi-layer filtering and verification approach for robust feature-to-code translation to ensure that the extraction functions run without crashing. Experimental evaluation results show that DeepFeature achieves an average AUROC improvement of 4.21-9.67% across eight diverse tasks compared to baseline methods. It outperforms state-of-the-art approaches on five tasks while maintaining comparable performance on the remaining tasks.

Hao Han, Jin Xie, Xuehao Ma et al.

Resolving real-world software engineering (SWE) issues with autonomous agents requires complex, long-horizon reasoning. Current pipelines are bottlenecked by unoptimized demonstration data, sparse execution rewards, and computationally prohibitive inference scaling, which collectively exacerbate token bloat, reward hacking, and policy degradation. We present SWE-TRACE (Trajectory Reduction and Agentic Criteria Evaluation), a unified framework optimizing the SWE agent lifecycle across data curation, reinforcement learning (RL), and test-time inference. First, we introduce an LLM multi-task cascading method, utilizing stepwise oracle verification to distill a 60K-instance Supervised Fine-Tuning (SFT) corpus strictly biased toward token-efficient, shortest-path trajectories. Second, to overcome the instability of sparse outcome rewards, we design a MemoryAugmented Agentic RL pipeline featuring a Rubric-Based Process Reward Model (PRM). An auxiliary Rubric-Agent provides dense, fine-grained heuristic feedback on intermediate steps, guiding the model through long-horizon tasks. Finally, we bridge training and inference by repurposing the PRM for heuristic-guided Test-Time Scaling (TTS). By dynamically evaluating and pruning action candidates at each step, SWE-TRACE achieves superior search efficiency without the latency overhead of standard parallel sampling. Extensive experiments on standard SWE benchmarks demonstrate that SWE-TRACE significantly advances the state-of-the-art, maximizing resolution rates while drastically reducing both token consumption and inference latency.

Sifan Li, Hongkai Chen, Yujun Cai et al.

Executable SQL generation is typically studied in text-to-SQL settings, where tables are provided as fully linearized textual schemas and contents. While effective, this formulation assumes access to structured text and incurs substantial token overhead, which is misaligned with many real-world scenarios where tables appear as visual artifacts in documents or webpages. We investigate whether compact optical representations can serve as an efficient interface for executable semantic parsing. We present OptiSQL, a vision-driven framework that generates executable SQL directly from table images and natural language questions using compact optical tokens. OptiSQL leverages an OCR-oriented visual encoder to compress table structure and content into a small set of optical tokens and fine-tunes a pretrained decoder for SQL generation while freezing the encoder to isolate representation sufficiency. Experiments on a visualized version of Spider 2.0-Snow show that OptiSQL retains strong execution accuracy while reducing table input tokens by an order of magnitude. Robustness analyses further demonstrate that optical tokens preserve essential structural information under visual perturbations.

Liyang Chen, Hongkai Chen, Yujun Cai et al.

Large Audio Language Models (LALMs) have demonstrated strong capabilities in audio understanding and reasoning. However, their performance on fine grained auditory perception remains unreliable, and existing approaches largely rely on data intensive training to internalize perceptual abilities. We propose AudioRouter, a reinforcement learning framework that enables LALMs to improve audio understanding by learning when and how to use external audio tools. Rather than tightly coupling tool usage with audio reasoning, AudioRouter formulates tool use as an explicit decision making problem and optimizes a lightweight routing policy while keeping the underlying reasoning model frozen. Experimental results show that AudioRouter achieves substantial improvements on standard audio understanding benchmarks while requiring up to 600x less training data to learn tool usage compared with conventional training paradigms. These findings suggest that learning effective tool usage offers a data efficient and scalable alternative to internalizing perceptual abilities in LALMs.

Zehua Zhang, Ati Priya Bajaj, Divij Handa et al.

Automatically compiling open-source software (OSS) projects is a vital, labor-intensive, and complex task, which makes it a good challenge for LLM Agents. Existing methods rely on manually curated rules and workflows, which cannot adapt to OSS that requires customized configuration or environment setup. Recent attempts using Large Language Models (LLMs) used selective evaluation on a subset of highly rated OSS, a practice that underestimates the realistic challenges of OSS compilation. In practice, compilation instructions are often absent, dependencies are undocumented, and successful builds may even require patching source files or modifying build scripts. We propose a more challenging and realistic benchmark, BUILD-BENCH, comprising OSS that are more diverse in quality, scale, and characteristics. Furthermore, we propose a strong baseline LLM-based agent, OSS-BUILD-AGENT, an effective system with enhanced build instruction retrieval module that achieves state-of-the-art performance on BUILD-BENCH and is adaptable to heterogeneous OSS characteristics. We also provide detailed analysis regarding different compilation method design choices and their influence to the whole task, offering insights to guide future advances. We believe performance on BUILD-BENCH can faithfully reflect an agent's ability to tackle compilation as a complex software engineering tasks, and, as such, our benchmark will spur innovation with a significant impact on downstream applications in the fields of software development and software security.

Bufang Yang, Siyang Jiang, Lilin Xu et al.

Large language models (LLMs) have the potential to transform digital healthcare, as evidenced by recent advances in LLM-based virtual doctors. However, current approaches rely on patient's subjective descriptions of symptoms, causing increased misdiagnosis. Recognizing the value of daily data from smart devices, we introduce a novel LLM-based multi-turn consultation virtual doctor system, DrHouse, which incorporates three significant contributions: 1) It utilizes sensor data from smart devices in the diagnosis process, enhancing accuracy and reliability. 2) DrHouse leverages continuously updating medical databases such as Up-to-Date and PubMed to ensure our model remains at diagnostic standard's forefront. 3) DrHouse introduces a novel diagnostic algorithm that concurrently evaluates potential diseases and their likelihood, facilitating more nuanced and informed medical assessments. Through multi-turn interactions, DrHouse determines the next steps, such as accessing daily data from smart devices or requesting in-lab tests, and progressively refines its diagnoses. Evaluations on three public datasets and our self-collected datasets show that DrHouse can achieve up to an 18.8% increase in diagnosis accuracy over the state-of-the-art baselines. The results of a 32-participant user study show that 75% medical experts and 91.7% patients are willing to use DrHouse.

Bufang Yang, Yunqi Guo, Lilin Xu et al.

Social interactions are fundamental to human life. The recent emergence of large language models (LLMs)-based virtual assistants has demonstrated their potential to revolutionize human interactions and lifestyles. However, existing assistive systems mainly provide reactive services to individual users, rather than offering in-situ assistance during live social interactions with conversational partners. In this study, we introduce SocialMind, the first LLM-based proactive AR social assistive system that provides users with in-situ social assistance. SocialMind employs human-like perception leveraging multi-modal sensors to extract both verbal and nonverbal cues, social factors, and implicit personas, incorporating these social cues into LLM reasoning for social suggestion generation. Additionally, SocialMind employs a multi-tier collaborative generation strategy and proactive update mechanism to display social suggestions on Augmented Reality (AR) glasses, ensuring that suggestions are timely provided to users without disrupting the natural flow of conversation. Evaluations on three public datasets and a user study with 20 participants show that SocialMind achieves 38.3% higher engagement compared to baselines, and 95% of participants are willing to use SocialMind in their live social interactions.

Bingliang Li, Fengyu Yang, Yuxin Mao et al.

Video-to-audio (V2A) generation utilizes visual-only video features to produce realistic sounds that correspond to the scene. However, current V2A models often lack fine-grained control over the generated audio, especially in terms of loudness variation and the incorporation of multi-modal conditions. To overcome these limitations, we introduce Tri-Ergon, a diffusion-based V2A model that incorporates textual, auditory, and pixel-level visual prompts to enable detailed and semantically rich audio synthesis. Additionally, we introduce Loudness Units relative to Full Scale (LUFS) embedding, which allows for precise manual control of the loudness changes over time for individual audio channels, enabling our model to effectively address the intricate correlation of video and audio in real-world Foley workflows. Tri-Ergon is capable of creating 44.1 kHz high-fidelity stereo audio clips of varying lengths up to 60 seconds, which significantly outperforms existing state-of-the-art V2A methods that typically generate mono audio for a fixed duration.

Hang Wu, Hongkai Chen, Yujun Cai et al.

Grounding natural language queries in graphical user interfaces (GUIs) poses unique challenges due to the diversity of visual elements, spatial clutter, and the ambiguity of language. In this paper, we introduce DiMo-GUI, a training-free framework for GUI grounding that leverages two core strategies: dynamic visual grounding and modality-aware optimization. Instead of treating the GUI as a monolithic image, our method splits the input into textual elements and iconic elements, allowing the model to reason over each modality independently using general-purpose vision-language models. When predictions are ambiguous or incorrect, DiMo-GUI dynamically focuses attention by generating candidate focal regions centered on the model's initial predictions and incrementally zooms into subregions to refine the grounding result. This hierarchical refinement process helps disambiguate visually crowded layouts without the need for additional training or annotations. We evaluate our approach on standard GUI grounding benchmarks and demonstrate consistent improvements over baseline inference pipelines, highlighting the effectiveness of combining modality separation with region-focused reasoning.

Wulin Xie, Yi-Fan Zhang, Chaoyou Fu et al. · pku

Existing MLLM benchmarks face significant challenges in evaluating Unified MLLMs (U-MLLMs) due to: 1) lack of standardized benchmarks for traditional tasks, leading to inconsistent comparisons; 2) absence of benchmarks for mixed-modality generation, which fails to assess multimodal reasoning capabilities. We present a comprehensive evaluation framework designed to systematically assess U-MLLMs. Our benchmark includes: Standardized Traditional Task Evaluation. We sample from 12 datasets, covering 10 tasks with 30 subtasks, ensuring consistent and fair comparisons across studies." 2. Unified Task Assessment. We introduce five novel tasks testing multimodal reasoning, including image editing, commonsense QA with image generation, and geometric reasoning. 3. Comprehensive Model Benchmarking. We evaluate 12 leading U-MLLMs, such as Janus-Pro, EMU3, VILA-U, and Gemini2-flash, alongside specialized understanding (e.g., Claude-3.5-Sonnet) and generation models (e.g., DALL-E-3). Our findings reveal substantial performance gaps in existing U-MLLMs, highlighting the need for more robust models capable of handling mixed-modality tasks effectively. The code and evaluation data can be found in https://mme-unify.github.io/.

Hongkai Chen, Zixin Luo, Yurun Tian et al.

Identifying robust and accurate correspondences across images is a fundamental problem in computer vision that enables various downstream tasks. Recent semi-dense matching methods emphasize the effectiveness of fusing relevant cross-view information through Transformer. In this paper, we propose several improvements upon this paradigm. Firstly, we introduce affine-based local attention to model cross-view deformations. Secondly, we present selective fusion to merge local and global messages from cross attention. Apart from network structure, we also identify the importance of enforcing spatial smoothness in loss design, which has been omitted by previous works. Based on these augmentations, our network demonstrate strong matching capacity under different settings. The full version of our network achieves state-of-the-art performance among semi-dense matching methods at a similar cost to LoFTR, while the slim version reaches LoFTR baseline's performance with only 15% computation cost and 18% parameters.

Heming Fu, Hongkai Chen, Shan Lin et al.

Alzheimer's Disease (AD) has become an increasingly critical global health concern, which necessitates effective monitoring solutions in smart health applications. However, the development of such solutions is significantly hindered by the scarcity of AD-specific activity datasets. To address this challenge, we propose SHADE-AD, a Large Language Model (LLM) framework for Synthesizing Human Activity Datasets Embedded with AD features. Leveraging both public datasets and our own collected data from 99 AD patients, SHADE-AD synthesizes human activity videos that specifically represent AD-related behaviors. By employing a three-stage training mechanism, it broadens the range of activities beyond those collected from limited deployment settings. We conducted comprehensive evaluations of the generated dataset, demonstrating significant improvements in downstream tasks such as Human Activity Recognition (HAR) detection, with enhancements of up to 79.69%. Detailed motion metrics between real and synthetic data show strong alignment, validating the realism and utility of the synthesized dataset. These results underscore SHADE-AD's potential to advance smart health applications by providing a cost-effective, privacy-preserving solution for AD monitoring.

Jinglin He, Yunqi Guo, Lai Kwan Lam et al.

Traditional Chinese Medicine (TCM) represents a rich repository of ancient medical knowledge that continues to play an important role in modern healthcare. Due to the complexity and breadth of the TCM literature, the integration of AI technologies is critical for its modernization and broader accessibility. However, this integration poses considerable challenges, including the interpretation of obscure classical Chinese texts and the modeling of intricate semantic relationships among TCM concepts. In this paper, we develop OpenTCM, an LLM-based system that combines a domain-specific TCM knowledge graph and Graph-based Retrieval-Augmented Generation (GraphRAG). First, we extract more than 3.73 million classical Chinese characters from 68 gynecological books in the Chinese Medical Classics Database, with the help of TCM and gynecology experts. Second, we construct a comprehensive multi-relational knowledge graph comprising more than 48,000 entities and 152,000 interrelationships, using customized prompts and Chinese-oriented LLMs such as DeepSeek and Kimi to ensure high-fidelity semantic understanding. Last, we empower OpenTCM with GraphRAG, enabling high-fidelity ingredient knowledge retrieval and diagnostic question-answering without model fine-tuning. Experimental evaluations demonstrate that OpenTCM achieves mean expert scores (MES) of 4.378 in ingredient information retrieval and 4.045 in diagnostic question-answering tasks, outperforming state-of-the-art solutions in real-world TCM use cases.

Chang Liu, Hongkai Chen, Yujun Cai et al.

Document understanding remains a significant challenge for multimodal large language models (MLLMs). While previous research has primarily focused on locating evidence pages through precise multimodal queries, our work investigates a fundamental yet overlooked aspect: how input format influences document comprehension performance. Through systematic analysis, we discover that raw OCR text often impairs rather than improves MLLMs' performance, which is a counterintuitive finding we attribute to attention dispersion and structure loss. To further substantiate our hypothesis, we propose a novel structure-preserving approach that encodes document elements using the LaTex paradigm, maintaining the hierarchical organization and spatial relationships critical for comprehension. Our attention analysis reveals that structured text induces structured attention patterns on both textual and visual content, directing models to focus on semantically meaningful regions while reducing attention waste. This approach significantly enhances MLLMs' document question answering performance across diverse document types without requiring architectural modifications or additional training.

Sifan Li, Hongkai Chen, Yujun Cai et al.

Vision Language Models (VLMs) have achieved impressive progress in multimodal reasoning; yet, they remain vulnerable to hallucinations, where outputs are not grounded in visual evidence. In this paper, we investigate a previously overlooked setting: logo hallucination, where models generate brand names or textual content despite logos containing no visible words. Using curated splits of pure symbols, hybrids, and text-bearing logos, as well as the challenging Hard-60 subset, we systematically measure hallucination across leading VLMs. We further probe robustness through nine structured perturbations and show that hallucinations persist even under strong distortions, with occlusion exposing the sharpest weaknesses. Embedding-level analysis with open-weight LLaVA demonstrates that hallucination is tied to a small subset of projector dimensions, and targeted ablation substantially reduces errors while preserving OCR accuracy. Together, these findings reveal that VLMs often rely on symbolic priors rather than genuine glyph perception, particularly for iconic circular logos, and that projector subspaces play a decisive role in this failure mode. Our work contributes both a novel diagnostic lens and actionable mitigation insights, highlighting projector disentanglement and OCR-guided decoding as promising directions for building more trustworthy multimodal systems.

Liyang Chen, Hongkai Chen, Yujun Cai et al.

Video-to-Audio generation has made remarkable strides in automatically synthesizing sound for video. However, existing evaluation metrics, which focus on semantic and temporal alignment, overlook a critical failure mode: models often generate acoustic events, particularly speech and music, that have no corresponding visual source. We term this phenomenon Insertion Hallucination and identify it as a systemic risk driven by dataset biases, such as the prevalence of off-screen sounds, that remains completely undetected by current metrics. To address this challenge, we first develop a systematic evaluation framework that employs a majority-voting ensemble of multiple audio event detectors. We also introduce two novel metrics to quantify the prevalence and severity of this issue: IH@vid (the fraction of videos with hallucinations) and IH@dur (the fraction of hallucinated duration). Building on this, we propose Posterior Feature Correction, a novel training-free inference-time method that mitigates IH. PFC operates in a two-pass process: it first generates an initial audio output to detect hallucinated segments, and then regenerates the audio after masking the corresponding video features at those timestamps. Experiments on several mainstream V2A benchmarks first reveal that state-of-the-art models suffer from severe IH. In contrast, our PFC method reduces both the prevalence and duration of hallucinations by over 50\% on average, without degrading, and in some cases even improving, conventional metrics for audio quality and temporal synchronization. Our work is the first to formally define, systematically measure, and effectively mitigate Insertion Hallucination, paving the way for more reliable and faithful V2A models.

Hang Wu, Yujun Cai, Haonan Ge et al.

Cinematography understanding refers to the ability to recognize not only the visual content of a scene but also the cinematic techniques that shape narrative meaning. This capability is attracting increasing attention, as it enhances multimodal understanding in real-world applications and underpins coherent content creation in film and media. As the most comprehensive benchmark for this task, ShotBench spans a wide range of cinematic concepts and VQA-style evaluations, with ShotVL achieving state-of-the-art results on it. However, our analysis reveals that ambiguous option design in ShotBench and ShotVL's shortcomings in reasoning consistency and instruction adherence undermine evaluation reliability, limiting fair comparison and hindering future progress. To overcome these issues, we systematically refine ShotBench through consistent option restructuring, conduct the first critical analysis of ShotVL's reasoning behavior, and introduce an extended evaluation protocol that jointly assesses task accuracy and core model competencies. These efforts lead to RefineShot, a refined and expanded benchmark that enables more reliable assessment and fosters future advances in cinematography understanding.

Siyang Jiang, Bufang Yang, Lilin Xu et al.

The rapid advancements in Large Vision Language Models (LVLMs) offer the potential to surpass conventional labeling by generating richer, more detailed descriptions of on-device human behavior understanding (HBU) in low-resolution vision systems, such as depth, thermal, and infrared. However, existing large vision language model (LVLM) approaches are unable to understand low-resolution data well as they are primarily designed for high-resolution data, such as RGB images. A quick fixing approach is to caption a large amount of low-resolution data, but it requires a significant amount of labor-intensive annotation efforts. In this paper, we propose a novel, labor-saving system, Llambda, designed to support low-resolution HBU. The core idea is to leverage limited labeled data and a large amount of unlabeled data to guide LLMs in generating informative captions, which can be combined with raw data to effectively fine-tune LVLM models for understanding low-resolution videos in HBU. First, we propose a Contrastive-Oriented Data Labeler, which can capture behavior-relevant information from long, low-resolution videos and generate high-quality pseudo labels for unlabeled data via contrastive learning. Second, we propose a Physical-Knowledge Guided Captioner, which utilizes spatial and temporal consistency checks to mitigate errors in pseudo labels. Therefore, it can improve LLMs' understanding of sequential data and then generate high-quality video captions. Finally, to ensure on-device deployability, we employ LoRA-based efficient fine-tuning to adapt LVLMs for low-resolution data. We evaluate Llambda using a region-scale real-world testbed and three distinct low-resolution datasets, and the experiments show that Llambda outperforms several state-of-the-art LVLM systems up to $40.03\%$ on average Bert-Score.

Ke Ma, Hongkai Chen, Shan Lin

Type 1 diabetes is a serious disease in which individuals are unable to regulate their blood glucose levels, leading to various medical complications. Artificial pancreas (AP) systems have been developed as a solution for type 1 diabetic patients to mimic the behavior of the pancreas and regulate blood glucose levels. However, current AP systems lack detection capabilities for exercise-induced glucose intake, which can last up to 4 to 8 hours. This incapability can lead to hypoglycemia, which if left untreated, could have serious consequences, including death. Existing exercise detection methods are either limited to single sensor data or use inaccurate models for exercise detection, making them less effective in practice. In this work, we propose an ensemble learning framework that combines a data-driven physiological model and a Siamese network to leverage multiple physiological signal streams for exercise detection with high accuracy. To evaluate the effectiveness of our proposed approach, we utilized a public dataset with 12 diabetic patients collected from an 8-week clinical trial. Our approach achieves a true positive rate for exercise detection of 86.4% and a true negative rate of 99.1%, outperforming state-of-the-art solutions.

Hongkai Chen, Zixin Luo, Jiahui Zhang et al.

Matching local features across images is a fundamental problem in computer vision. Targeting towards high accuracy and efficiency, we propose Seeded Graph Matching Network, a graph neural network with sparse structure to reduce redundant connectivity and learn compact representation. The network consists of 1) Seeding Module, which initializes the matching by generating a small set of reliable matches as seeds. 2) Seeded Graph Neural Network, which utilizes seed matches to pass messages within/across images and predicts assignment costs. Three novel operations are proposed as basic elements for message passing: 1) Attentional Pooling, which aggregates keypoint features within the image to seed matches. 2) Seed Filtering, which enhances seed features and exchanges messages across images. 3) Attentional Unpooling, which propagates seed features back to original keypoints. Experiments show that our method reduces computational and memory complexity significantly compared with typical attention-based networks while competitive or higher performance is achieved.

Xuyang Bai, Zixin Luo, Lei Zhou et al.

Removing outlier correspondences is one of the critical steps for successful feature-based point cloud registration. Despite the increasing popularity of introducing deep learning methods in this field, spatial consistency, which is essentially established by a Euclidean transformation between point clouds, has received almost no individual attention in existing learning frameworks. In this paper, we present PointDSC, a novel deep neural network that explicitly incorporates spatial consistency for pruning outlier correspondences. First, we propose a nonlocal feature aggregation module, weighted by both feature and spatial coherence, for feature embedding of the input correspondences. Second, we formulate a differentiable spectral matching module, supervised by pairwise spatial compatibility, to estimate the inlier confidence of each correspondence from the embedded features. With modest computation cost, our method outperforms the state-of-the-art hand-crafted and learning-based outlier rejection approaches on several real-world datasets by a significant margin. We also show its wide applicability by combining PointDSC with different 3D local descriptors.

Zixin Luo, Lei Zhou, Xuyang Bai et al.

This work focuses on mitigating two limitations in the joint learning of local feature detectors and descriptors. First, the ability to estimate the local shape (scale, orientation, etc.) of feature points is often neglected during dense feature extraction, while the shape-awareness is crucial to acquire stronger geometric invariance. Second, the localization accuracy of detected keypoints is not sufficient to reliably recover camera geometry, which has become the bottleneck in tasks such as 3D reconstruction. In this paper, we present ASLFeat, with three light-weight yet effective modifications to mitigate above issues. First, we resort to deformable convolutional networks to densely estimate and apply local transformation. Second, we take advantage of the inherent feature hierarchy to restore spatial resolution and low-level details for accurate keypoint localization. Finally, we use a peakiness measurement to relate feature responses and derive more indicative detection scores. The effect of each modification is thoroughly studied, and the evaluation is extensively conducted across a variety of practical scenarios. State-of-the-art results are reported that demonstrate the superiority of our methods.

Hongkai Chen, Nicola Paoletti, Scott A. Smolka et al.

Even though model predictive control (MPC) is currently the main algorithm for insulin control in the artificial pancreas (AP), it usually requires complex online optimizations, which are infeasible for resource-constrained medical devices. MPC also typically relies on state estimation, an error-prone process. In this paper, we introduce a novel approach to AP control that uses Imitation Learning to synthesize neural-network insulin policies from MPC-computed demonstrations. Such policies are computationally efficient and, by instrumenting MPC at training time with full state information, they can directly map measurements into optimal therapy decisions, thus bypassing state estimation. We apply Bayesian inference via Monte Carlo Dropout to learn policies, which allows us to quantify prediction uncertainty and thereby derive safer therapy decisions. We show that our control policies trained under a specific patient model readily generalize (in terms of model parameters and disturbance distributions) to patient cohorts, consistently outperforming traditional MPC with state estimation.