In this paper, we consider the problem of long-term point tracking, which requires consistent identification of points across video frames under significant appearance changes, motion, and occlusion. We target the online setting, i.e. tracking points frame-by-frame, making it suitable for real-time and streaming applications. We extend our prior model Track-On into Track-On2, a simple and efficient transformer-based model for online long-term tracking. Track-On2 improves both performance and efficiency through architectural refinements, more effective use of memory, and improved synthetic training strategies. Unlike prior approaches that rely on full-sequence access or iterative updates, our model processes frames causally and maintains temporal coherence via a memory mechanism, which is key to handling drift and occlusions without requiring future frames. At inference, we perform coarse patch-level classification followed by refinement. Beyond architecture, we systematically study synthetic training setups and their impact on memory behavior, showing how they shape temporal robustness over long sequences. Through comprehensive experiments, Track-On2 achieves state-of-the-art results across five synthetic and real-world benchmarks, surpassing prior online trackers and even strong offline methods that exploit bidirectional context. These results highlight the effectiveness of causal, memory-based architectures trained purely on synthetic data as scalable solutions for real-world point tracking. Project page: https://kuis-ai.github.io/track_on2

Robot-assisted surgery (RAS) has become a critical paradigm in modern surgery, promoting patient recovery and reducing the burden on surgeons through minimally invasive approaches. To fully realize its potential, however, a precise understanding of the visual data generated during surgical procedures is essential. Previous studies have predominantly focused on single-task approaches, but real surgical scenes involve complex temporal dynamics and diverse instrument interactions that limit comprehensive understanding. Moreover, the effective application of multi-task learning (MTL) requires sufficient pixel-level segmentation data, which are difficult to obtain due to the high cost and expertise required for annotation. In particular, long-term annotations such as phases and steps are available for every frame, whereas short-term annotations such as surgical instrument segmentation and action detection are provided only for key frames, resulting in a significant temporal-spatial imbalance. To address these challenges, we propose a novel framework that combines optical flow-based segmentation label interpolation with multi-task learning. optical flow estimated from annotated key frames is used to propagate labels to adjacent unlabeled frames, thereby enriching sparse spatial supervision and balancing temporal and spatial information for training. This integration improves both the accuracy and efficiency of surgical scene understanding and, in turn, enhances the utility of RAS.

Dexterous robotic hands enable robots to perform complex manipulations that require fine-grained control and adaptability. Achieving such manipulation is challenging because the high degrees of freedom tightly couple hand and arm motions, making learning and control difficult. Successful dexterous manipulation relies not only on precise hand motions, but also on accurate spatial positioning of the arm and coordinated arm-hand dynamics. However, most existing visuomotor policies represent arm and hand actions in a single combined space, which often causes high-dimensional hand actions to dominate the coupled action space and compromise arm control. To address this, we propose DQ-RISE, which quantizes hand states to simplify hand motion prediction while preserving essential patterns, and applies a continuous relaxation that allows arm actions to diffuse jointly with these compact hand states. This design enables the policy to learn arm-hand coordination from data while preventing hand actions from overwhelming the action space. Experiments show that DQ-RISE achieves more balanced and efficient learning, paving the way toward structured and generalizable dexterous manipulation. Project website: http://rise-policy.github.io/DQ-RISE/

Many manipulation tasks require memory beyond the current observation, yet most visuomotor policies rely on the Markov assumption and thus struggle with repeated states or long-horizon dependencies. Existing methods attempt to extend observation horizons but remain insufficient for diverse memory requirements. To this end, we propose an object-centric history representation based on point tracking, which abstracts past observations into a compact and structured form that retains only essential task-relevant information. Tracked points are encoded and aggregated at the object level, yielding a compact history representation that can be seamlessly integrated into various visuomotor policies. Our design provides full history-awareness with high computational efficiency, leading to improved overall task performance and decision accuracy. Through extensive evaluations on diverse manipulation tasks, we show that our method addresses multiple facets of memory requirements - such as task stage identification, spatial memorization, and action counting, as well as longer-term demands like continuous and pre-loaded memory - and consistently outperforms both Markovian baselines and prior history-based approaches. Project website: http://tonyfang.net/history

Multimodal desire understanding, a task closely related to both emotion and sentiment that aims to infer human intentions from visual and textual cues, is an emerging yet underexplored task in affective computing with applications in social media analysis. Existing methods for related tasks predominantly focus on mining verbal cues, often overlooking the effective utilization of non-verbal cues embedded in images. To bridge this gap, we propose a Symmetrical Bidirectional Multimodal Learning Framework for Desire, Emotion, and Sentiment Recognition (SyDES). The core of SyDES is to achieve bidirectional fine-grained modal alignment between text and image modalities. Specifically, we introduce a mixed-scaled image strategy that combines global context from low-resolution images with fine-grained local features via masked image modeling (MIM) on high-resolution sub-images, effectively capturing intention-related visual representations. Then, we devise symmetrical cross-modal decoders, including a text-guided image decoder and an image-guided text decoder, which enable mutual reconstruction and refinement between modalities, facilitating deep cross-modal interaction. Furthermore, a set of dedicated loss functions is designed to harmonize potential conflicts between the MIM and modal alignment objectives during optimization. Extensive evaluations on the MSED benchmark demonstrate the superiority of our approach, which establishes a new state-of-the-art performance with 1.1% F1-score improvement in desire understanding. Consistent gains in emotion and sentiment recognition further validate its generalization ability and the necessity of utilizing non-verbal cues. Our code is available at: https://github.com/especiallyW/SyDES.

Artificial intelligence has shown significant promise in chest radiography, where deep learning models can approach radiologist-level diagnostic performance. Progress has been accelerated by large public datasets such as MIMIC-CXR, ChestX-ray14, PadChest, and CheXpert, which provide hundreds of thousands of labelled images with pathology annotations. However, these datasets also present important limitations. Automated label extraction from radiology reports introduces errors, particularly in handling uncertainty and negation, and radiologist review frequently disagrees with assigned labels. In addition, domain shift and population bias restrict model generalisability, while evaluation practices often overlook clinically meaningful measures. We conduct a systematic analysis of these challenges, focusing on label quality, dataset bias, and domain shift. Our cross-dataset domain shift evaluation across multiple model architectures revealed substantial external performance degradation, with pronounced reductions in AUPRC and F1 scores relative to internal testing. To assess dataset bias, we trained a source-classification model that distinguished datasets with near-perfect accuracy, and performed subgroup analyses showing reduced performance for minority age and sex groups. Finally, expert review by two board-certified radiologists identified significant disagreement with public dataset labels. Our findings highlight important clinical weaknesses of current benchmarks and emphasise the need for clinician-validated datasets and fairer evaluation frameworks.

Comparing vision language models on videos is particularly complex, as the performances is jointly determined by the model's visual representation capacity and the frame-sampling strategy used to construct the input. Current video benchmarks are suspected to suffer from substantial frame-sampling bias, as models are evaluated with different frame selection strategies. In this work, we propose the first frame-accurate benchmark of state-of-the-art small VLMs for video question-answering, evaluated under controlled frame-sampling strategies. Our results confirm the suspected bias and highlight both data-specific and task-specific behaviors of SVLMs under different frame-sampling techniques. By open-sourcing our benchmarking code, we provide the community with a reproducible and unbiased protocol for evaluating video VLMs and emphasize the need for standardized frame-sampling strategies tailored to each benchmarking dataset in future research.

This paper introduces a new approach to generating sample paths of unknown Markovian stochastic differential equations (SDEs) using diffusion models, a class of generative AI methods commonly employed in image and video applications. Unlike the traditional Monte Carlo methods for simulating SDEs, which require explicit specifications of the drift and diffusion coefficients, ours takes a model-free, data-driven approach. Given a finite set of sample paths from an SDE, we utilize conditional diffusion models to generate new, synthetic paths of the same SDE. Numerical experiments show that our method consistently outperforms two alternative methods in terms of the Kullback--Leibler (KL) divergence between the distributions of the target SDE paths and the generated ones. Moreover, we present a theoretical error analysis deriving an explicit bound on the said KL divergence. Finally, in simulation and empirical studies, we leverage these synthetically generated sample paths to boost the performance of reinforcement learning algorithms for continuous-time mean--variance portfolio selection, hinting promising applications of our study in financial analysis and decision-making.

Modeling human-object interactions (HOI) from an egocentric perspective is a critical yet challenging task, particularly when relying on sparse signals from wearable devices like smart glasses and watches. We present ECHO, the first unified framework to jointly recover human pose, object motion, and contact dynamics solely from head and wrist tracking. To tackle the underconstrained nature of this problem, we introduce a novel tri-variate diffusion process with independent noise schedules that models the mutual dependencies between the human, object, and interaction modalities. This formulation allows ECHO to operate with flexible input configurations, making it robust to intermittent tracking and capable of leveraging partial observations. Crucially, it enables training on a combination of large-scale human motion datasets and smaller HOI collections, learning strong priors while capturing interaction nuances. Furthermore, we employ a smooth inpainting inference mechanism that enables the generation of temporally consistent interactions for arbitrarily long sequences. Extensive evaluations demonstrate that ECHO achieves state-of-the-art performance, significantly outperforming existing methods lacking such flexibility.

Natural disasters pose significant challenges to timely and accurate damage assessment due to their sudden onset and the extensive areas they affect. Traditional assessment methods are often labor-intensive, costly, and hazardous to personnel, making them impractical for rapid response, especially in resource-limited settings. This study proposes a novel, cost-effective framework that leverages aerial drone footage, an advanced AI-based video super-resolution model, Video Restoration Transformer (VRT), and Gemma3:27b, a 27 billion parameter Visual Language Model (VLM). This integrated system is designed to improve low-resolution disaster footage, identify structural damage, and classify buildings into four damage categories, ranging from no/slight damage to total destruction, along with associated risk levels. The methodology was validated using pre- and post-event drone imagery from the 2023 Turkey earthquakes (courtesy of The Guardian) and satellite data from the 2013 Moore Tornado (xBD dataset). The framework achieved a classification accuracy of 84.5%, demonstrating its ability to provide highly accurate results. Furthermore, the system's accessibility allows non-technical users to perform preliminary analyses, thereby improving the responsiveness and efficiency of disaster management efforts.

While reinforcement learning (RL) has proven highly effective for general reasoning in vision-language models, its application to tasks requiring deep understanding of information-rich images and structured output generation remains underexplored. Chart-to-code generation exemplifies this challenge, demanding complex reasoning over visual charts to produce structured code. Supervised fine-tuning (SFT) alone is often insufficient, highlighting the need for effective RL strategies tailored to structured outputs. In this paper, we systematically investigate the performance plateau of SFT through large-scale experiments and propose Multimodal Structured Reinforcement Learning (MSRL) for chart-to-code generation. We construct the largest training corpus to date, with 3 million chart-code pairs curated from real-world tables in arXiv papers, addressing the limitations of previous synthetic datasets. Despite achieving state-of-the-art performance, our experiments show that simply increasing SFT data eventually leads to diminishing improvements. To break this plateau, MSRL employs a multi-granularity reward system that integrates both textual and visual feedback. At the textual level, rule-based rewards validate fine-grained code details, while at the visual level, a model-based reward assesses the structural similarity between rendered code and ground-truth charts. We implement a two-stage curriculum training strategy, first optimizing the model with textual rewards and then incorporating visual signals for further enhancement. Experimental results demonstrate that MSRL substantially breaks the SFT plateau, improving high-level metrics by 6.2% and 9.9% on ChartMimic and ReachQA benchmarks, respectively. Notably, our method outperforms all existing approaches in the chart domain and achieves competitive results with advanced closed-source models.

Fleets of autonomous robots are increasingly deployed with multiple human operators in communication-restricted environments for exploration and intervention tasks such as subterranean inspection, reconnaissance, and search-and-rescue. In these settings, communication is often limited to short-range ad-hoc links, making it difficult to coordinate exploration while supporting online human-fleet interactions. Existing work on multi-robot exploration largely focuses on information gathering itself, but pays limited attention to the fact that operators and robots issue time-critical requests during execution. These requests may require different communication structures, ranging from intermittent status delivery to sustained video streaming and teleoperation. To address this challenge, this paper presents MoRoCo, an online topology-adaptive framework for multi-operator multi-robot coordination under restricted communication. MoRoCo is built on a latency-bounded intermittent communication backbone that guarantees a prescribed delay for information collected by any robot to reach an operator, together with a detach-and-rejoin mechanism that enables online team resizing and topology reconfiguration. On top of this backbone, the framework instantiates request-consistent communication subgraphs to realize different modes of operator-robot interaction by jointly assigning robot roles, positions, and communication topology. It further supports the online decomposition and composition of these subgraphs using only local communication, allowing multiple requests to be serviced during exploration. The framework extends to heterogeneous fleets, multiple teams, and robot failures. Extensive human-in-the-loop simulations and hardware experiments demonstrate effective and reliable coordination under restricted communication.

In this paper, we describe an implementation of the two-phase image segmentation algorithm proposed by Goldstein, Bresson, Osher in \cite{gold:bre}. This algorithm partitions the domain of a given 2d image into foreground and background regions, and each pixel of the image is assigned membership to one of these two regions. The underlying assumption for the segmentation model is that the pixel values of the input image can be summarized by two distinct average values, and that the region boundaries are smooth. Accordingly, the model is defined as an energy in which the variable is a region membership function to assign pixels to either region, originally proposed by Chan and Vese in \cite{chan:vese}. This energy is the sum of image data terms in the regions and a length penalty for region boundaries. Goldstein, Bresson, Osher modify the energy of Chan-Vese in \cite{gold:bre} so that their new energy can be minimized efficiently using the split Bregman method to produce an equivalent two-phase segmentation. We provide a detailed implementation of this method \cite{gold:bre}, and document its performance with several images over a range of algorithm parameters.

Memes, as a widely used mode of online communication, often serve as vehicles for spreading harmful content. However, limitations in data accessibility and the high costs of dataset curation hinder the development of robust meme moderation systems. To address this challenge, in this work, we introduce a first-of-its-kind dataset - TOXICTAGS consisting of 6,300 real-world meme-based posts annotated in two stages: (i) binary classification into toxic and normal, and (ii) fine-grained labelling of toxic memes as hateful, dangerous, or offensive. A key feature of this dataset is that it is enriched with auxiliary metadata of socially relevant tags, enhancing the context of each meme. In addition, we propose a novel entropy guided multi-tasking framework - STEMTOX - that integrates the generation of socially grounded tags with a robust classification framework. Experimental results show that incorporating these tags substantially enhances the performance of state-of-the-art VLMs in toxicity detection tasks. Our contributions offer a novel and scalable foundation for improved content moderation in multimodal online environments. Warning: Contains potentially toxic contents.

Foundation models for 3D vision have recently demonstrated remarkable capabilities in 3D perception. However, extending these models to large-scale RGB stream 3D reconstruction remains challenging due to memory limitations. In this work, we propose VGGT-Long, a simple yet effective system that pushes the limits of monocular 3D reconstruction to kilometer-scale, unbounded outdoor environments. Our approach addresses the scalability bottlenecks of existing models through a chunk-based processing strategy combined with overlapping alignment and lightweight loop closure optimization. Without requiring camera calibration, depth supervision or model retraining, VGGT-Long achieves trajectory and reconstruction performance comparable to traditional methods. We evaluate our method on KITTI, Waymo, and Virtual KITTI datasets. VGGT-Long not only runs successfully on long RGB sequences where foundation models typically fail, but also produces accurate and consistent geometry across various conditions. Our results highlight the potential of leveraging foundation models for scalable monocular 3D scene in real-world settings, especially for autonomous driving scenarios. Code is available at https://github.com/DengKaiCQ/VGGT-Long.

Story points are unitless, project-specific effort estimates that help developers plan their sprints. Traditionally, developers have collaboratively estimated story points using planning poker or other manual techniques. Machine learning can reduce this burden, but only with sufficient context from the historical decisions made by the project team. That is, state-of-the-art models, such as GPT2SP and FastText-SVM, only make accurate (within-project) predictions when they are trained on data from the same project. The goal of this study is to streamline story point estimation by evaluating a comparative learning-based framework for calibrating project-specific story point prediction models. Instead of assigning a specific story point value to every backlog item, developers are presented with pairs of items and asked to indicate which item requires more effort. Using these comparative judgments, a machine learning model was trained to predict the story point estimates. We empirically evaluated our technique using data from 23,313 manual estimates across 16 projects. The model trained on comparative judgments achieved, on average, a 0.34 Spearman's rank correlation coefficient between its predictions and the ground truth story points. This is similar to, if not better than, the performance of a state-of-the-art regression model trained on ground truth story points. Through human subject experiments, the advantages of comparative judgments were validated - higher confidence, lower annotation time, and comparable agreement were observed for comparative judgments compared to direct ratings. In summary, the proposed comparative learning approach is more efficient than regression-based approaches, given its better performance, lower required annotation time, and higher training data reliability.

Many program synthesis tasks prove too challenging for even state-of-the-art language models to solve in single attempts. Search-based evolutionary methods offer a promising alternative by exploring solution spaces iteratively, but their effectiveness remain limited by the fixed capabilities of the underlying generative model. We propose SOAR, a method that learns program synthesis by integrating language models into a self-improving evolutionary loop. SOAR alternates between (1) an evolutionary search that uses an LLM to sample and refine candidate solutions, and (2) a hindsight learning phase that converts search attempts into valid problem-solution pairs used to fine-tune the LLM's sampling and refinement capabilities\, -- \,enabling increasingly effective search in subsequent iterations. On the challenging ARC-AGI benchmark, SOAR achieves significant performance gains across model scales and iterations, leveraging positive transfer between the sampling and refinement finetuning tasks. These improvements carry over to test-time adaptation, enabling SOAR to solve 52\% of the public test set. Our code is open-sourced at: https://github.com/flowersteam/SOAR

We develop a novel framework for uncertainty quantification in operator learning, the Stochastic Operator Network (SON). SON combines the stochastic optimal control concepts of the Stochastic Neural Network (SNN) with the DeepONet. By formulating the branch net as an SDE and backpropagating through the adjoint BSDE, we replace the gradient of the loss function with the gradient of the Hamiltonian from Stohastic Maximum Principle in the SGD update. This allows SON to learn the uncertainty present in operators through its diffusion parameters. We then demonstrate the effectiveness of SON when replicating several noisy operators in 2D and 3D.

Large vision-language models (LVLMs) have demonstrated remarkable capabilities by integrating pre-trained vision encoders with large language models (LLMs). Similar to single-modal LLMs, chain-of-thought (CoT) prompting has been adapted for LVLMs to enhance multi-modal reasoning by generating intermediate rationales based on visual and textual inputs. While CoT is assumed to improve grounding and accuracy in LVLMs, our experiments reveal a key challenge: existing LVLMs often ignore the contents of generated rationales in CoT reasoning. To address this, we re-formulate multi-modal CoT reasoning as a KL-constrained reward maximization focused on rationale-conditional log-likelihood. As the optimal solution, we propose rationale-enhanced decoding (RED), a novel plug-and-play inference-time decoding strategy. RED harmonizes visual and rationale information by multiplying distinct image-conditional and rationale-conditional next token distributions. Extensive experiments show that RED consistently and significantly improves reasoning over standard CoT and other decoding methods across multiple benchmarks and LVLMs. Our work offers a practical and effective approach to improve both the faithfulness and accuracy of CoT reasoning in LVLMs, paving the way for more reliable rationale-grounded multi-modal systems. Code is available at https://github.com/yshinya6/red/.

Humans can imagine and manipulate visual images mentally, a capability known as spatial visualization. While many multi-modal benchmarks assess reasoning on visible visual information, the ability to infer unseen relationships through spatial visualization remains insufficiently evaluated as a spatial skill. This reliance on publicly sourced problems from IQ tests or math competitions risks data contamination and compromises assessment reliability. To this end, we introduce SpatialViz-Bench, a comprehensive multi-modal benchmark for spatial visualization with 12 tasks across 4 sub-abilities, comprising 1,180 programmatically generated problems, a scalable framework that allows for expansion to ensure fair and continuously reliable evaluations. Our evaluation of 27 Multi-modal Large Language Models (MLLMs) reveals wide performance variations, demonstrates the benchmark's strong discriminative power, and uncovers counter-intuitive findings: Chain-of-Thought (CoT) prompting paradoxically degrades accuracy on open-source models. Through statistical and qualitative analysis of error types, SpatialViz-Bench demonstrates that state-of-the-art MLLMs exhibit deficiencies in spatial visualization tasks, thereby addressing a significant lacuna in the field. The benchmark data and evaluation code are publicly available.

Large Vision-Language Models (LVLMs) face a tug-of-war between powerful linguistic priors and visual evidence, often leading to \emph{semantic drift}: a progressive detachment from the input image that can abruptly emerge at specific decoding steps. Through a token-level diagnosis, we show that hallucination is frequently triggered not by the absence of grounded candidates, but by a failure of selection -- the model chooses a linguistically convenient yet visually unfaithful token even when better grounded alternatives exist. Motivated by this insight, we propose \textbf{D}ynamic \textbf{L}ogits \textbf{C}alibration (DLC), a training-free decoding framework that introduces a lightweight visual referee to intervene exactly when drift happens. At each step, DLC performs a dual-aspect grounding check on top-$k$ candidates: (1) it assesses the intrinsic visual relevance of a candidate token and (2) its contextual visual coherence. These signals are evaluated against an adaptive historical baseline to compute a relative visual advantage, which is then used to dynamically calibrate logits and favor grounded tokens. Extensive experiments on CHAIR, POPE, SHR, GPT-4o evaluation, and MME demonstrate that DLC consistently reduces hallucinations across multiple LVLMs while preserving response quality. Further analyses validate robustness to different vision backbones and demonstrate a favorable trade-off between output quality and computational cost as the candidate pool size varies. Code will be released on https://github.com/JiaheChen2002/DLC.

Reconstructing animatable 3D humans from casually captured images of articulated subjects without camera or pose information is highly practical but remains challenging due to view misalignment, occlusions, and the absence of structural priors. In this work, we present LHM++, an efficient large-scale human reconstruction model that generates high-quality, animatable 3D avatars within seconds from one or multiple pose-free images. At its core is an Encoder-Decoder Point-Image Transformer architecture that progressively encodes and decodes 3D geometric point features to improve efficiency, while fusing hierarchical 3D point features with image features through multimodal attention. The fused features are decoded into 3D Gaussian splats to recover detailed geometry and appearance. To further enhance visual fidelity, we introduce a lightweight 3D-aware neural animation renderer that refines the rendering quality of reconstructed avatars in real time. Extensive experiments show that our method produces high-fidelity, animatable 3D humans without requiring camera or pose annotations. Our code and project page are available at https://lingtengqiu.github.io/LHM++/

We present BIS Reasoning 1.0, the first large-scale Japanese dataset of syllogistic reasoning problems explicitly designed to evaluate belief-inconsistent reasoning in large language models (LLMs). Unlike prior resources such as NeuBAROCO and JFLD, which emphasize general or belief-aligned logic, BIS Reasoning 1.0 systematically introduces logically valid yet belief-inconsistent syllogisms to expose belief bias, the tendency to accept believable conclusions irrespective of validity. We benchmark a representative suite of cutting-edge models, including OpenAI GPT-5 variants, GPT-4o, Qwen, and prominent Japanese LLMs, under a uniform, zero-shot protocol. Reasoning-centric models achieve near-perfect accuracy on BIS Reasoning 1.0 (e.g., Qwen3-32B $\approx$99% and GPT-5-mini up to $\approx$99.7%), while GPT-4o attains around 80%. Earlier Japanese-specialized models underperform, often well below 60%, whereas the latest llm-jp-3.1-13b-instruct4 markedly improves to the mid-80% range. These results indicate that robustness to belief-inconsistent inputs is driven more by explicit reasoning optimization than by language specialization or scale alone. Our analysis further shows that even top-tier systems falter when logical validity conflicts with intuitive or factual beliefs, and that performance is sensitive to prompt design and inference-time reasoning effort. We discuss implications for safety-critical domains, including law, healthcare, and scientific literature, where strict logical fidelity must override intuitive belief to ensure reliability.

We aim to improve the reasoning capabilities of language models via reinforcement learning (RL). Recent RL post-trained models like DeepSeek-R1 have demonstrated reasoning abilities on mathematical and coding tasks. However, prior studies suggest that using RL alone to improve reasoning on inherently difficult tasks is less effective. Here, we draw inspiration from curriculum learning and propose to schedule tasks from easy to hard (E2H), allowing LLMs to build reasoning skills gradually. Our method is termed E2H Reasoner. Empirically, we observe that, although easy tasks are important initially, fading them out through appropriate scheduling is essential in preventing overfitting. Theoretically, we establish convergence guarantees for E2H Reasoner within an approximate policy iteration framework. We derive finite-sample complexity bounds and show that when tasks are appropriately decomposed and conditioned, learning through curriculum stages requires fewer total samples than direct learning. Experiments across multiple domains show that E2H Reasoner significantly improves the reasoning ability of small LLMs (1.5B to 3B), which otherwise struggle when trained with vanilla RL alone, highlighting the effectiveness of our method. Our code can be found on https://github.com/divelab/E2H-Reasoning.

Recent advances in video understanding have been driven by MLLMs. But these MLLMs are good at analyzing short videos, while suffering from difficulties in understanding videos with a longer context. To address this difficulty, several agent methods have been proposed, using MLLMs as agents for retrieving extra contextual knowledge in a long video. However, most existing agents ignore the key fact that a long video is composed with multiple shots, i.e., to answer the user question from a long video, it is critical to deeply understand its relevant shots like human. Without such insight, these agents often mistakenly find redundant even noisy temporal context, restricting their capacity for long video understanding. To fill this gap, we propose VideoChat-A1, a novel long video agent paradigm. Different from the previous works, our VideoChat-A1 can deeply think with long videos, via a distinct chain-of-shot reasoning paradigm. More specifically, it can progressively select the relevant shots of user question, and look into these shots in a coarse-to-fine partition. By multi-modal reasoning along the shot chain, VideoChat-A1 can effectively mimic step-by-step human thinking process, allowing the interactive discovery of preferable temporal context for thoughtful understanding in long videos. Extensive experiments show that, VideoChat-A1 achieves the state-of-the-art performance on the mainstream long video QA benchmarks, e.g., it achieves 77.0 on VideoMME (w/ subs) and 70.1 on EgoSchema, outperforming its strong baselines (e.g., InternVL2.5-8B and InternVideo2.5-8B), by up to 10.1\% and 6.2\%. Compared to leading closed-source GPT-4o and Gemini 1.5 Pro, VideoChat-A1 offers competitive accuracy, but only with 7% input frames and 12% inference time on average. The code is available on https://github.com/SpXace/VideoChat-A1.

Speech inherently contains rich acoustic information that extends far beyond the textual language. In real-world spoken language understanding, effective interpretation often requires integrating semantic meaning (e.g., content), paralinguistic features (e.g., emotions, speed, pitch) and phonological characteristics (e.g., prosody, intonation, rhythm), which are embedded in speech. While recent multimodal Speech Large Language Models (SpeechLLMs) have demonstrated remarkable capabilities in processing audio information, their ability to perform fine-grained perception and complex reasoning in natural speech remains largely unexplored. To address this gap, we introduce MMSU, a comprehensive benchmark designed specifically for understanding and reasoning in spoken language. MMSU comprises 5,000 meticulously curated audio-question-answer triplets across 47 distinct tasks. To ground our benchmark in linguistic theory, we systematically incorporate a wide range of linguistic phenomena, including phonetics, prosody, rhetoric, syntactics, semantics, and paralinguistics. Through a rigorous evaluation of 14 advanced SpeechLLMs, we identify substantial room for improvement in existing models, highlighting meaningful directions for future optimization. MMSU establishes a new standard for comprehensive assessment of spoken language understanding, providing valuable insights for developing more sophisticated human-AI speech interaction systems. MMSU benchmark is available at https://huggingface.co/datasets/ddwang2000/MMSU. Evaluation Code is available at https://github.com/dingdongwang/MMSU.

Detecting structural change in dynamic network data has wide-ranging applications. Existing approaches typically divide the data into time bins, extract network features within each bin, and then compare these features over time. This introduces an inherent tradeoff between temporal resolution and the statistical stability of the extracted features. Despite this tradeoff, reminiscent of time-frequency tradeoffs in signal processing, most methods rely on a fixed temporal resolution. Choosing an appropriate resolution parameter is typically difficult and can be especially problematic in domains like cybersecurity, where anomalous behavior may emerge at multiple time scales. We address this challenge by proposing ANIE (Adaptive Network Intensity Estimation), a multi-resolution framework designed to automatically identify the time scales at which network structure evolves, enabling the joint detection of both rapid and gradual changes. Modeling interactions as Poisson processes, our method proceeds in two steps: (1) estimating a low-dimensional subspace of node behavior, and (2) deriving a set of novel empirical affinity coefficients that quantify change in interaction intensity between latent factors and support statistical testing for structural change across time scales. We provide theoretical guarantees for subspace estimation and the asymptotic behavior of the affinity coefficients, enabling model-based change detection. Experiments on synthetic networks show that ANIE adapts to the appropriate time resolution and is able to capture sharp structural changes while remaining robust to noise. Furthermore, applications to real-world data showcase the practical benefits of ANIE's multiresolution approach to detecting structural change over fixed resolution methods.

Federated learning (FL) enables collaborative model training across distributed edge devices while preserving data privacy, and typically operates in a round-based synchronous manner. However, synchronous FL suffers from latency bottlenecks due to device heterogeneity, where slower clients (stragglers) delay or degrade global updates. Prior solutions, such as fixed deadlines, client selection, and layer-wise partial aggregation, alleviate the effect of stragglers, but treat round timing and local workload as static parameters, limiting their effectiveness under strict time constraints. We propose ADEL-FL, a novel framework that jointly optimizes per-round deadlines and user-specific batch sizes for layer-wise aggregation. Our approach formulates a constrained optimization problem minimizing the expected L2 distance to the global optimum under total training time and global rounds. We provide a convergence analysis under exponential compute models and prove that ADEL-FL yields unbiased updates with bounded variance. Extensive experiments demonstrate that ADEL-FL outperforms alternative methods in both convergence rate and final accuracy under heterogeneous conditions.

Object removal requires eliminating not only the target object but also its associated visual effects such as shadows and reflections. However, diffusion-based inpainting and removal methods often introduce artifacts, hallucinate contents, alter background, and struggle to remove object effects accurately. To address these challenges, we propose ObjectClear, a novel framework that decouples foreground removal from background reconstruction via an adaptive target-aware attention mechanism. This design empowers the model to precisely localize and remove both objects and their effects while maintaining high background fidelity. Moreover, the learned attention maps are leveraged for an attention-guided fusion strategy during inference, further enhancing visual consistency. To facilitate the training and evaluation, we construct OBER, a large-scale dataset for OBject-Effect Removal, which provides paired images with and without object-effects, along with precise masks for both objects and their effects. The dataset comprises high-quality captured and simulated data, covering diverse objects, effects, and complex multi-object scenes. Extensive experiments demonstrate that ObjectClear outperforms prior methods, achieving superior object-effect removal quality and background fidelity, especially in challenging scenarios.

Leveraging multimodal large models for image segmentation has become a prominent research direction. However, existing approaches typically rely heavily on manually annotated datasets that include explicit reasoning processes, which are costly and time-consuming to produce. Recent advances suggest that reinforcement learning (RL) can endow large models with reasoning capabilities without requiring such reasoning-annotated data. In this paper, we propose SAM-R1, a novel framework that enables multimodal large models to perform fine-grained reasoning in image understanding tasks. Our approach is the first to incorporate fine-grained segmentation settings during the training of multimodal reasoning models. By integrating task-specific, fine-grained rewards with a tailored optimization objective, we further enhance the model's reasoning and segmentation alignment. We also leverage the Segment Anything Model (SAM) as a strong and flexible reward provider to guide the learning process. With only 3k training samples, SAM-R1 achieves strong performance across multiple benchmarks, demonstrating the effectiveness of reinforcement learning in equipping multimodal models with segmentation-oriented reasoning capabilities.