Yibo Liu

Xiang Yue, Yuansheng Ni, Kai Zhang et al. · microsoft-research, princeton

We introduce MMMU: a new benchmark designed to evaluate multimodal models on massive multi-discipline tasks demanding college-level subject knowledge and deliberate reasoning. MMMU includes 11.5K meticulously collected multimodal questions from college exams, quizzes, and textbooks, covering six core disciplines: Art & Design, Business, Science, Health & Medicine, Humanities & Social Science, and Tech & Engineering. These questions span 30 subjects and 183 subfields, comprising 30 highly heterogeneous image types, such as charts, diagrams, maps, tables, music sheets, and chemical structures. Unlike existing benchmarks, MMMU focuses on advanced perception and reasoning with domain-specific knowledge, challenging models to perform tasks akin to those faced by experts. The evaluation of 14 open-source LMMs as well as the proprietary GPT-4V(ision) and Gemini highlights the substantial challenges posed by MMMU. Even the advanced GPT-4V and Gemini Ultra only achieve accuracies of 56% and 59% respectively, indicating significant room for improvement. We believe MMMU will stimulate the community to build next-generation multimodal foundation models towards expert artificial general intelligence.

Ningyuan Huang, Yash R. Deshpande, Yibo Liu et al. · amazon-science

We propose a method to make natural language understanding models more parameter efficient by storing knowledge in an external knowledge graph (KG) and retrieving from this KG using a dense index. Given (possibly multilingual) downstream task data, e.g., sentences in German, we retrieve entities from the KG and use their multimodal representations to improve downstream task performance. We use the recently released VisualSem KG as our external knowledge repository, which covers a subset of Wikipedia and WordNet entities, and compare a mix of tuple-based and graph-based algorithms to learn entity and relation representations that are grounded on the KG multimodal information. We demonstrate the usefulness of the learned entity representations on two downstream tasks, and show improved performance on the multilingual named entity recognition task by $0.3\%$--$0.7\%$ F1, while we achieve up to $2.5\%$ improvement in accuracy on the visual sense disambiguation task. All our code and data are available in: \url{https://github.com/iacercalixto/visualsem-kg}.

Kimi Team, Tongtong Bai, Yifan Bai et al.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel agent orchestration framework that dynamically decomposes complex tasks into heterogeneous sub-problems and executes them concurrently. Extensive evaluations show that Kimi K2.5 achieves state-of-the-art results across various domains including coding, vision, reasoning, and agentic tasks. Agent Swarm also reduces latency by up to $4.5\times$ over single-agent baselines. We release the post-trained Kimi K2.5 model checkpoint to facilitate future research and real-world applications of agentic intelligence.

Yibo Liu, Hunter Schofield, Jinjun Shan

The fiducial marker system for LiDAR is crucial for the robotic application but it is still rare to date. In this paper, an Intensity Image-based LiDAR Fiducial Marker (IILFM) system is developed. This system only requires an unstructured point cloud with intensity as the input and it has no restriction on marker placement and shape. A marker detection method that locates the predefined 3D fiducials in the point cloud through the intensity image is introduced. Then, an approach that utilizes the detected 3D fiducials to estimate the LiDAR 6-DOF pose that describes the transmission from the world coordinate system to the LiDAR coordinate system is developed. Moreover, all these processes run in real-time (approx 40 Hz on Livox Mid-40 and approx 143 Hz on VLP-16). Qualitative and quantitative experiments are conducted to demonstrate that the proposed system has similar convenience and accuracy as the conventional visual fiducial marker system. The codes and results are available at: https://github.com/York-SDCNLab/IILFM.

Yibo Liu, Jinjun Shan, Hunter Schofield

The LiDAR fiducial tag, akin to the well-known AprilTag used in camera applications, serves as a convenient resource to impart artificial features to the LiDAR sensor, facilitating robotics applications. Unfortunately, the existing LiDAR fiducial tag localization methods do not apply to 3D LiDAR maps while resolving this problem is beneficial to LiDAR-based relocalization and navigation. In this paper, we develop a novel approach to directly localize fiducial tags on a 3D LiDAR prior map, returning the tag poses (labeled by ID number) and vertex locations (labeled by index) w.r.t. the global coordinate system of the map. In particular, considering that fiducial tags are thin sheet objects indistinguishable from the attached planes, we design a new pipeline that gradually analyzes the 3D point cloud of the map from the intensity and geometry perspectives, extracting potential tag-containing point clusters. Then, we introduce an intermediate-plane-based method to further check if each potential cluster has a tag and compute the vertex locations and tag pose if found. We conduct both qualitative and quantitative experiments to demonstrate that our approach is the first method applicable to localize tags on a 3D LiDAR map while achieving better accuracy compared to previous methods. The open-source implementation of this work is available at: https://github.com/York-SDCNLab/Marker-Detection-General.

Yue Wu, Yibo Liu, Maoguo Gong et al.

Multi-view point cloud registration is fundamental in 3D reconstruction. Since there are close connections between point clouds captured from different viewpoints, registration performance can be enhanced if these connections be harnessed properly. Therefore, this paper models the registration problem as multi-task optimization, and proposes a novel bi-channel knowledge sharing mechanism for effective and efficient problem solving. The modeling of multi-view point cloud registration as multi-task optimization are twofold. By simultaneously considering the local accuracy of two point clouds as well as the global consistency posed by all the point clouds involved, a fitness function with an adaptive threshold is derived. Also a framework of the co-evolutionary search process is defined for the concurrent optimization of multiple fitness functions belonging to related tasks. To enhance solution quality and convergence speed, the proposed bi-channel knowledge sharing mechanism plays its role. The intra-task knowledge sharing introduces aiding tasks that are much simpler to solve, and useful information is shared across aiding tasks and the original tasks, accelerating the search process. The inter-task knowledge sharing explores commonalities buried among the original tasks, aiming to prevent tasks from getting stuck to local optima. Comprehensive experiments conducted on model object as well as scene point clouds show the efficacy of the proposed method.

Wu Yue, Peiran Gong, Maoguo Gong et al.

Point cloud registration (PCR) is a popular research topic in computer vision. Recently, the registration method in an evolutionary way has received continuous attention because of its robustness to the initial pose and flexibility in objective function design. However, most evolving registration methods cannot tackle the local optimum well and they have rarely investigated the success ratio, which implies the probability of not falling into local optima and is closely related to the practicality of the algorithm. Evolutionary multi-task optimization (EMTO) is a widely used paradigm, which can boost exploration capability through knowledge transfer among related tasks. Inspired by this concept, this study proposes a novel evolving registration algorithm via EMTO, where the multi-task configuration is based on the idea of solution space cutting. Concretely, one task searching in cut space assists another task with complex function landscape in escaping from local optima and enhancing successful registration ratio. To reduce unnecessary computational cost, a sparse-to-dense strategy is proposed. In addition, a novel fitness function robust to various overlap rates as well as a problem-specific metric of computational cost is introduced. Compared with 8 evolving approaches, 4 traditional approaches and 3 deep learning approaches on the object-scale and scene-scale registration datasets, experimental results demonstrate that the proposed method has superior performances in terms of precision and tackling local optima.

Hulin Wang, Zion Leonahenahe Basque, Jie Hu et al.

Recent LLM-based systems have made automated vulnerability repair increasingly practical, but two challenges remain. First, without strong signals about where a bug originates, repair agents drift toward shallow edits that silence the observed failure while leaving the underlying defect unresolved. Second, finding the root cause for bugs is hard: even developers familiar with the codebase frequently produce fixes that address symptoms rather than the root cause, and LLM-based agents, operating with noisier context and less program understanding, are no exception. We present Kumushi, a root-cause-driven patching agent that addresses both challenges by combining diversified dynamic fault localization with evidence-weighted ranking to focus the LLM on the code most relevant to the defect. To rigorously measure whether Kumushi produces genuinely better patches, we also introduce a two-tier patch quality metric that pairs automated oracle validation with structured expert assessment of patches. Evaluated on 178 C/C++ vulnerabilities, Kumushi substantially outperforms prior specialized repair agents under automated evaluation while matching a frontier commercial coding agent. Expert assessment then reveals differences that oracles cannot: Kumushi produces more root-cause fixes and fewer superficial patches, and is preferred in the majority of decisive pairwise comparisons. Together, these results demonstrate that progress in automated vulnerability repair requires not only stronger patching systems, but also richer evaluation methods capable of distinguishing genuine fixes from oracle-passing ones.

Kimi Team, Yu Zhang, Zongyu Lin et al.

We introduce Kimi Linear, a hybrid linear attention architecture that, for the first time, outperforms full attention under fair comparisons across various scenarios -- including short-context, long-context, and reinforcement learning (RL) scaling regimes. At its core lies Kimi Delta Attention (KDA), an expressive linear attention module that extends Gated DeltaNet with a finer-grained gating mechanism, enabling more effective use of limited finite-state RNN memory. Our bespoke chunkwise algorithm achieves high hardware efficiency through a specialized variant of the Diagonal-Plus-Low-Rank (DPLR) transition matrices, which substantially reduces computation compared to the general DPLR formulation while remaining more consistent with the classical delta rule. We pretrain a Kimi Linear model with 3B activated parameters and 48B total parameters, based on a layerwise hybrid of KDA and Multi-Head Latent Attention (MLA). Our experiments show that with an identical training recipe, Kimi Linear outperforms full MLA with a sizeable margin across all evaluated tasks, while reducing KV cache usage by up to 75% and achieving up to 6 times decoding throughput for a 1M context. These results demonstrate that Kimi Linear can be a drop-in replacement for full attention architectures with superior performance and efficiency, including tasks with longer input and output lengths. To support further research, we open-source the KDA kernel and vLLM implementations, and release the pre-trained and instruction-tuned model checkpoints.

Jinsheng Guo, Zhenhao Weng, Yibo Liu et al.

Graph fraud detection has long depended on Graph Neural Networks (GNNs) to propagate and aggregate information across relational data. A critical obstacle in practice, however, is that fraudsters frequently disguise themselves by forging numerous connections with benign users, causing fraud signals to be progressively diluted during neighborhood aggregation and undermining detection reliability. While recent efforts have used Large Language Models (LLMs) to provide rich semantic cues for fraud detection, the underlying intent behind suspicious connections remains insufficiently explored. Compounding this issue, the scarcity of annotated fraud samples makes it difficult to train detectors that remain robust under heavy camouflage. To address these gaps, we propose L2IR, an LLM-driven Latent Intent Revealing framework for graph fraud detection. By uncovering latent intent from both user behaviors and suspicious connections, L2IR extracts intent-aware representations from raw behavioral traces and reasons about the true purpose behind individual connections, effectively distinguishing supportive links from misleading ones. It further incorporates adaptive self-training to enhance robustness under limited supervision. Evaluations on two real-world datasets characterized by pervasive camouflage demonstrate that L2IR surpasses strong baselines and can function as a plug-in enhancement for a range of GNN-based detectors, improving AUPRC by up to 8.27%.

Kimi Team, Angang Du, Bohong Yin et al. · pku, tsinghua

We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking-2506. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), the latest model exhibits strong long-horizon reasoning capabilities (64.0 on MMMU, 46.3 on MMMU-Pro, 56.9 on MathVision, 80.1 on MathVista, 65.2 on VideoMMMU) while obtaining robust general abilities. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.

Jingyuan Liu, Jianlin Su, Xingcheng Yao et al.

Recently, the Muon optimizer based on matrix orthogonalization has demonstrated strong results in training small-scale language models, but the scalability to larger models has not been proven. We identify two crucial techniques for scaling up Muon: (1) adding weight decay and (2) carefully adjusting the per-parameter update scale. These techniques allow Muon to work out-of-the-box on large-scale training without the need of hyper-parameter tuning. Scaling law experiments indicate that Muon achieves $\sim\!2\times$ computational efficiency compared to AdamW with compute optimal training. Based on these improvements, we introduce Moonlight, a 3B/16B-parameter Mixture-of-Expert (MoE) model trained with 5.7T tokens using Muon. Our model improves the current Pareto frontier, achieving better performance with much fewer training FLOPs compared to prior models. We open-source our distributed Muon implementation that is memory optimal and communication efficient. We also release the pretrained, instruction-tuned, and intermediate checkpoints to support future research.

Yibo Liu, Kelly Zhu, Guile Wu et al.

Reconstructing 3D vehicles from noisy and sparse partial point clouds is of great significance to autonomous driving. Most existing 3D reconstruction methods cannot be directly applied to this problem because they are elaborately designed to deal with dense inputs with trivial noise. In this work, we propose a novel framework, dubbed MV-DeepSDF, which estimates the optimal Signed Distance Function (SDF) shape representation from multi-sweep point clouds to reconstruct vehicles in the wild. Although there have been some SDF-based implicit modeling methods, they only focus on single-view-based reconstruction, resulting in low fidelity. In contrast, we first analyze multi-sweep consistency and complementarity in the latent feature space and propose to transform the implicit space shape estimation problem into an element-to-set feature extraction problem. Then, we devise a new architecture to extract individual element-level representations and aggregate them to generate a set-level predicted latent code. This set-level latent code is an expression of the optimal 3D shape in the implicit space, and can be subsequently decoded to a continuous SDF of the vehicle. In this way, our approach learns consistent and complementary information among multi-sweeps for 3D vehicle reconstruction. We conduct thorough experiments on two real-world autonomous driving datasets (Waymo and KITTI) to demonstrate the superiority of our approach over state-of-the-art alternative methods both qualitatively and quantitatively.

Yibo Liu, Zheyuan Yang, Guile Wu et al.

Generating 3D vehicle assets from in-the-wild observations is crucial to autonomous driving. Existing image-to-3D methods cannot well address this problem because they learn generation merely from image RGB information without a deeper understanding of in-the-wild vehicles (such as car models, manufacturers, etc.). This leads to their poor zero-shot prediction capability to handle real-world observations with occlusion or tricky viewing angles. To solve this problem, in this work, we propose VQA-Diff, a novel framework that leverages in-the-wild vehicle images to create photorealistic 3D vehicle assets for autonomous driving. VQA-Diff exploits the real-world knowledge inherited from the Large Language Model in the Visual Question Answering (VQA) model for robust zero-shot prediction and the rich image prior knowledge in the Diffusion model for structure and appearance generation. In particular, we utilize a multi-expert Diffusion Models strategy to generate the structure information and employ a subject-driven structure-controlled generation mechanism to model appearance information. As a result, without the necessity to learn from a large-scale image-to-3D vehicle dataset collected from the real world, VQA-Diff still has a robust zero-shot image-to-novel-view generation ability. We conduct experiments on various datasets, including Pascal 3D+, Waymo, and Objaverse, to demonstrate that VQA-Diff outperforms existing state-of-the-art methods both qualitatively and quantitatively.

Junqing He, Kunhao Pan, Xiaoqun Dong et al.

While large language models (LLMs) are equipped with longer text input capabilities than before, they are struggling to seek correct information in long contexts. The "lost in the middle" problem challenges most LLMs, referring to the dramatic decline in accuracy when correct information is located in the middle. To overcome this crucial issue, this paper proposes to enhance the information searching and reflection ability of LLMs in long contexts via specially designed tasks called Attention Strengthening Multi-doc QA (ASM QA). Following these tasks, our model excels in focusing more precisely on the desired information. Experimental results show substantial improvement in Multi-doc QA and other benchmarks, superior to state-of-the-art models by 13.7% absolute gain in shuffled settings, by 21.5% in passage retrieval task. We release our model, Ziya-Reader to promote related research in the community.

Kimi Team, Yifan Bai, Yiping Bao et al. · tsinghua

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

Zheyuan Yang, Yibo Liu, Guile Wu et al.

We present a solution for 3D object generation of ICCV 2023 OmniObject3D Challenge. In recent years, 3D object generation has made great process and achieved promising results, but it remains a challenging task due to the difficulty of generating complex, textured, and high-fidelity results. To resolve this problem, we study learning effective NeRFs and SDFs representations with 3D Generative Adversarial Networks (GANs) for 3D object generation. Specifically, inspired by recent works, we use the efficient geometry-aware 3D GANs as the backbone incorporating with label embedding and color mapping, which enables to train the model on different taxonomies simultaneously. Then, through a decoder, we aggregate the resulting features to generate Neural Radiance Fields (NeRFs) based representations for rendering high-fidelity synthetic images. Meanwhile, we optimize Signed Distance Functions (SDFs) to effectively represent objects with 3D meshes. Besides, we observe that this model can be effectively trained with only a few images of each object from a variety of classes, instead of using a great number of images per object or training one model per class. With this pipeline, we can optimize an effective model for 3D object generation. This solution is among the top 3 in the ICCV 2023 OmniObject3D Challenge.

Kimi Team, Angang Du, Bofei Gao et al. · pku, tsinghua

Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).

Zonghan Yang, Shengjie Wang, Kelin Fu et al.

Large Language Models (LLMs) are increasingly applied to software engineering (SWE), with SWE-bench as a key benchmark. Solutions are split into SWE-Agent frameworks with multi-turn interactions and workflow-based Agentless methods with single-turn verifiable steps. We argue these paradigms are not mutually exclusive: reasoning-intensive Agentless training induces skill priors, including localization, code edit, and self-reflection that enable efficient and effective SWE-Agent adaptation. In this work, we first curate the Agentless training recipe and present Kimi-Dev, an open-source SWE LLM achieving 60.4\% on SWE-bench Verified, the best among workflow approaches. With additional SFT adaptation on 5k publicly-available trajectories, Kimi-Dev powers SWE-Agents to 48.6\% pass@1, on par with that of Claude 3.5 Sonnet (241022 version). These results show that structured skill priors from Agentless training can bridge workflow and agentic frameworks for transferable coding agents.

Yibo Liu, Stanko Oparnica, Simon Shewchun-Jakaitis et al.

Contact-rich assembly is fundamental in robotics but poses significant challenges due to uncertainties in relative poses, such as misalignments and small clearances in peg-in-hole tasks. Existing approaches typically address search and high-precision insertion separately, because these tasks involve distinct action patterns. However, supporting both tasks within a single model, without switching models or weights, is desirable for intelligent assembly systems. In this work, we propose SI-Diff, a framework that learns both search and high-precision insertion through a force-domain diffusion policy. To this end, we introduce a new mode-conditioning mechanism that enables the policy to capture distinct action behaviors under a single framework. Moreover, we develop a new search teacher policy that can generate diverse trajectories. By training on successful and efficient demonstrations provided by the teacher policy, the model learns the mapping from tactile and end-effector velocity observations to effective action behaviors. We conduct thorough experiments to show that SI-Diff extends the tolerance to x-y misalignments from 2 mm to 5 mm compared to the state-of-the-art baseline, TacDiffusion, while also demonstrating strong zero-shot transferability to unseen shapes.

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.

Yibo Liu, Amaldev Haridevan, Hunter Schofield et al.

Feature extraction or localization based on the fiducial marker could fail due to motion blur in real-world robotic applications. To solve this problem, a lightweight generative adversarial network, named Ghost-DeblurGAN, for real-time motion deblurring is developed in this paper. Furthermore, on account that there is no existing deblurring benchmark for such task, a new large-scale dataset, YorkTag, is proposed that provides pairs of sharp/blurred images containing fiducial markers. With the proposed model trained and tested on YorkTag, it is demonstrated that when applied along with fiducial marker systems to motion-blurred images, Ghost-DeblurGAN improves the marker detection significantly. The datasets and codes used in this paper are available at: https://github.com/York-SDCNLab/Ghost-DeblurGAN.

Houda Alberts, Teresa Huang, Yash Deshpande et al.

An exciting frontier in natural language understanding (NLU) and generation (NLG) calls for (vision-and-) language models that can efficiently access external structured knowledge repositories. However, many existing knowledge bases only cover limited domains, or suffer from noisy data, and most of all are typically hard to integrate into neural language pipelines. To fill this gap, we release VisualSem: a high-quality knowledge graph (KG) which includes nodes with multilingual glosses, multiple illustrative images, and visually relevant relations. We also release a neural multi-modal retrieval model that can use images or sentences as inputs and retrieves entities in the KG. This multi-modal retrieval model can be integrated into any (neural network) model pipeline. We encourage the research community to use VisualSem for data augmentation and/or as a source of grounding, among other possible uses. VisualSem as well as the multi-modal retrieval models are publicly available and can be downloaded in this URL: https://github.com/iacercalixto/visualsem

Yingwei Ma, Rongyu Cao, Yongchang Cao et al.

Recent advancements in LLM-based agents have led to significant progress in automatic software engineering, particularly in software maintenance and evolution. Despite these encouraging advances, current research faces two major challenges. First, SOTA performance primarily depends on closed-source models, which significantly limits the technology's accessibility, and potential for customization in diverse SE tasks. Second, these models are predominantly trained on static code data, lacking a deep understanding of the dynamic interactions, iterative problem-solving processes, and evolutionary characteristics inherent in software development. To address these challenges, our study adopts a software engineering perspective. We recognize that real-world software maintenance and evolution processes encompass not only static code data but also developers' thought processes, utilization of external tools, and the interaction between different functional personnel. Consequently, we introduce the Lingma SWE-GPT series, comprising Lingma SWE-GPT 7B and 72B. By learning from and simulating real-world code submission activities, Lingma SWE-GPT systematically incorporates the dynamic interactions and iterative problem-solving inherent in software development process, thereby achieving a more comprehensive understanding of software improvement processes. We conducted experimental evaluations using SWE-bench Verified benchmark. The results demonstrate that Lingma SWE-GPT 72B successfully resolves 30.20% of the GitHub issues, marking a significant improvement in automatic issue resolution (22.76% relative improvement compared to Llama 3.1 405B), approaching the performance of closed-source models (31.80\% issues of GPT-4o resolved). Notably, Lingma SWE-GPT 7B resolves 18.20% of the issues, highlighting the potential for applying smaller models to ASE tasks.

Lanshan He, Haozhou Pang, Qi Gan et al.

Cutscenes are carefully choreographed cinematic sequences embedded in video games and interactive media, serving as the primary vehicle for narrative delivery, character development, and emotional engagement. Producing cutscenes is inherently complex: it demands seamless coordination across screenwriting, cinematography, character animation, voice acting, and technical direction, often requiring days to weeks of collaborative effort from multidisciplinary teams to produce minutes of polished content. In this work, we present Cutscene Agent, an LLM agent framework for automated end-to-end cutscene generation. The framework makes three contributions: (1)~a Cutscene Toolkit built on the Model Context Protocol (MCP) that establishes \emph{bidirectional} integration between LLM agents and the game engine -- agents not only invoke engine operations but continuously observe real-time scene state, enabling closed-loop generation of editable engine-native cinematic assets; (2)~a multi-agent system where a director agent orchestrates specialist subagents for animation, cinematography, and sound design, augmented by a visual reasoning feedback loop for perception-driven refinement; and (3)~CutsceneBench, a hierarchical evaluation benchmark for cutscene generation. Unlike typical tool-use benchmarks that evaluate short, isolated function calls, cutscene generation requires long-horizon, multi-step orchestration of dozens of interdependent tool invocations with strict ordering constraints -- a capability dimension that existing benchmarks do not cover. We evaluate a range of LLMs on CutsceneBench and analyze their performance across this challenging task.

Yuan Ren, Guile Wu, Runhao Li et al.

Urban scene reconstruction is crucial for real-world autonomous driving simulators. Although existing methods have achieved photorealistic reconstruction, they mostly focus on pinhole cameras and neglect fisheye cameras. In fact, how to effectively simulate fisheye cameras in driving scene remains an unsolved problem. In this work, we propose UniGaussian, a novel approach that learns a unified 3D Gaussian representation from multiple camera models for urban scene reconstruction in autonomous driving. Our contributions are two-fold. First, we propose a new differentiable rendering method that distorts 3D Gaussians using a series of affine transformations tailored to fisheye camera models. This addresses the compatibility issue of 3D Gaussian splatting with fisheye cameras, which is hindered by light ray distortion caused by lenses or mirrors. Besides, our method maintains real-time rendering while ensuring differentiability. Second, built on the differentiable rendering method, we design a new framework that learns a unified Gaussian representation from multiple camera models. By applying affine transformations to adapt different camera models and regularizing the shared Gaussians with supervision from different modalities, our framework learns a unified 3D Gaussian representation with input data from multiple sources and achieves holistic driving scene understanding. As a result, our approach models multiple sensors (pinhole and fisheye cameras) and modalities (depth, semantic, normal and LiDAR point clouds). Our experiments show that our method achieves superior rendering quality and fast rendering speed for driving scene simulation.

Ning Yang, Chengzhi Wang, Yibo Liu et al.

Large Language Models (LLMs) in long-context scenarios are severely constrained by the linear growth of Key-Value (KV) cache memory. Existing KV compression methods rely either on static thresholds and attention-only heuristics or on coarse memory budget allocation. Under tight memory budgets, these methods overlook two key factors: prompt-dependent variation in compression risk and functional heterogeneity across attention heads, which destabilize token selection and lead to tail failures. To address these challenges, we propose CompilerKV, a risk-adaptive and head-aware compression framework that compiles offline experience into reusable decision tables for prefill-only deployment. CompilerKV integrates two key synergistic components: (i) a Head Heterogeneity Table, learned via offline contextual bandits, which assigns head-specific reliability weights to govern functional differences across attention heads explicitly; and (ii) a Risk-Adaptive Threshold Gating mechanism that jointly models attention entropy and local perplexity, transforming prompt-level risk into deployable retention thresholds. Experiments on LongBench show CompilerKV dominates SOTA methods under a 512-token budget, recovering 97.7\% of FullKV performance while achieving up to +5.2 points gain over the strongest competitor.

Yibo Liu, Zhaodong Jiang, Binbin Xu et al.

This work focuses on model-free zero-shot 6D object pose estimation for robotics applications. While existing methods can estimate the precise 6D pose of objects, they heavily rely on curated CAD models or reference images, the preparation of which is a time-consuming and labor-intensive process. Moreover, in real-world scenarios, 3D models or reference images may not be available in advance and instant robot reaction is desired. In this work, we propose a novel framework named HIPPo, which eliminates the need for curated CAD models and reference images by harnessing image-to-3D priors from Diffusion Models, enabling model-free zero-shot 6D pose estimation. Specifically, we construct HIPPo Dreamer, a rapid image-to-mesh model built on a multiview Diffusion Model and a 3D reconstruction foundation model. Our HIPPo Dreamer can generate a 3D mesh of any unseen objects from a single glance in just a few seconds. Then, as more observations are acquired, we propose to continuously refine the diffusion prior mesh model by joint optimization of object geometry and appearance. This is achieved by a measurement-guided scheme that gradually replaces the plausible diffusion priors with more reliable online observations. Consequently, HIPPo can instantly estimate and track the 6D pose of a novel object and maintain a complete mesh for immediate robotic applications. Thorough experiments on various benchmarks show that HIPPo outperforms state-of-the-art methods in 6D object pose estimation when prior reference images are limited.

Yibo Liu, Liam Shatzel, Brandon Haworth et al.

Animating and simulating crowds using an agent-based approach is a well-established area where every agent in the crowd is individually controlled such that global human-like behaviour emerges. We observe that human navigation and movement in crowds are often influenced by complex social and environmental interactions, driven mainly by language and dialogue. However, most existing work does not consider these dimensions and leads to animations where agent-agent and agent-environment interactions are largely limited to steering and fixed higher-level goal extrapolation. We propose a novel method that exploits large language models (LLMs) to control agents' movement. Our method has two main components: a dialogue system and language-driven navigation. We periodically query agent-centric LLMs conditioned on character personalities, roles, desires, and relationships to control the generation of inter-agent dialogue when necessitated by the spatial and social relationships with neighbouring agents. We then use the conversation and each agent's personality, emotional state, vision, and physical state to control the navigation and steering of each agent. Our model thus enables agents to make motion decisions based on both their perceptual inputs and the ongoing dialogue. We validate our method in two complex scenarios that exemplify the interplay between social interactions, steering, and crowding. In these scenarios, we observe that grouping and ungrouping of agents automatically occur. Additionally, our experiments show that our method serves as an information-passing mechanism within the crowd. As a result, our framework produces more realistic crowd simulations, with emergent group behaviours arising naturally from any environmental setting.

Team Seedance, Heyi Chen, Siyan Chen et al.

Recent strides in video generation have paved the way for unified audio-visual generation. In this work, we present Seedance 1.5 pro, a foundational model engineered specifically for native, joint audio-video generation. Leveraging a dual-branch Diffusion Transformer architecture, the model integrates a cross-modal joint module with a specialized multi-stage data pipeline, achieving exceptional audio-visual synchronization and superior generation quality. To ensure practical utility, we implement meticulous post-training optimizations, including Supervised Fine-Tuning (SFT) on high-quality datasets and Reinforcement Learning from Human Feedback (RLHF) with multi-dimensional reward models. Furthermore, we introduce an acceleration framework that boosts inference speed by over 10X. Seedance 1.5 pro distinguishes itself through precise multilingual and dialect lip-syncing, dynamic cinematic camera control, and enhanced narrative coherence, positioning it as a robust engine for professional-grade content creation. Seedance 1.5 pro is now accessible on Volcano Engine at https://console.volcengine.com/ark/region:ark+cn-beijing/experience/vision?type=GenVideo.

Zishuo Bao, Yibo Liu, Changyutao Qiu

With the rise of different language model architecture, fine-tuning is becoming even more important for down stream tasks Model gets messy, finding proper hyperparameters for fine-tuning. Although BO has been tried for hyperparameter tuning, most of the existing methods are oblivious to the fact that BO relies on careful choices of acquisition functions, which are essential components of BO that guide how much to explore versus exploit during the optimization process; Different acquisition functions have different levels of sensitivity towards training loss and validation performance; existing methods often just apply an acquisition function no matter if the training and validation performance are sensitive to the acquisition function or not. This work introduces{Bilevel - BO - SWA}, a model fusion approach coupled with a bilevel BO strategy to improve the fine - tunning of large language models. Our work on mixture of acquisition functions like EI and UCB into nested opt loops, where inner loop perform minimization of training loss while outer loops optimized w.r.t. val metric. Experiments on GLUE tasks using RoBERTA - base show that when using EI and UCB, there is an improvement in generalization, and fine - tuning can be improved by up to 2.7%.

Ning Yang, Hai Lin, Yibo Liu et al.

Ensuring that large language models (LLMs) generate outputs aligned with human preferences is important for safe and effective AI interactions. While Direct Preference Optimization (DPO) employs an implicit reward function to optimize the policy model, however, it and its related variants overlook the differential importance of individual tokens and are sensitive to judgment noise in preference datasets during generation. Although recent methods attempt to assess the important weight of tokens via probability prediction or simplistic weighting schemes, these evaluation methods are prone to biases and still cannot fully address these issues. To solve this problem, we propose the Token-Importance Guided Direct Preference Optimization (TI-DPO), which introduces two key innovations: the gradient-based token-importance weights that dynamically prioritize critical tokens, and a triple loss that explicitly guides model outputs to approach human-preferred responses and stay away from non-preferred responses. Experimental results show that TI-DPO achieves higher accuracy and stronger generative diversity, providing more stable and computationally efficient solutions compared with DPO and other RLHF methods.

Yibo Liu, Zhixin Fang, Sune Darkner et al.

We propose mesh-free fluid simulations that exploit a kinematic neural basis for velocity fields represented by an MLP. We design a set of losses that ensures that these neural bases approximate fundamental physical properties such as orthogonality, divergence-free, boundary alignment, and smoothness. Our neural bases can then be used to fit an input sketch of a flow, which will inherit the same fundamental properties from the bases. We then can animate such flow in real-time using standard time integrators. Our neural bases can accommodate different domains, moving boundaries, and naturally extend to three dimensions.

Yibo Liu

LiDAR sensors are essential for autonomous systems, yet LiDAR fiducial markers (LFMs) lag behind visual fiducial markers (VFMs) in adoption and utility. Bridging this gap is vital for robotics and computer vision but challenging due to the sparse, unstructured nature of 3D LiDAR data and 2D-focused fiducial marker designs. This dissertation proposes a novel framework for mapping and localization using LFMs is proposed to benefit a variety of real-world applications, including the collection of 3D assets and training data for point cloud registration, 3D map merging, Augmented Reality (AR), and many more. First, an Intensity Image-based LiDAR Fiducial Marker (IFM) system is introduced, using thin, letter-sized markers compatible with VFMs. A detection method locates 3D fiducials from intensity images, enabling LiDAR pose estimation. Second, an enhanced algorithm extends detection to 3D maps, increasing marker range and facilitating tasks like 3D map merging. This method leverages both intensity and geometry, overcoming limitations of geometry-only detection approaches. Third, a new LFM-based mapping and localization method registers unordered, low-overlap point clouds. It employs adaptive threshold detection and a two-level graph framework to solve a maximum a-posteriori (MAP) problem, optimizing point cloud and marker poses. Additionally, the Livox-3DMatch dataset is introduced, improving learning-based multiview point cloud registration methods. Extensive experiments with various LiDAR models in diverse indoor and outdoor scenes demonstrate the effectiveness and superiority of the proposed framework.

Yibo Liu, Jinjun Shan, Amaldev Haridevan et al.

Point cloud registration is a prerequisite for many applications in computer vision and robotics. Most existing methods focus on pairwise registration of two point clouds with high overlap. Although there have been some methods for low overlap cases, they struggle in degraded scenarios. This paper introduces a novel framework dubbed L-PR, designed to register unordered low overlap multiview point clouds leveraging LiDAR fiducial markers. We refer to them as LiDAR fiducial markers, but they are the same as the popular AprilTag and ArUco markers, thin sheets of paper that do not affect the 3D geometry of the environment. We first propose an improved adaptive threshold marker detection method to provide robust detection results when the viewpoints among point clouds change dramatically. Then, we formulate the unordered multiview point cloud registration problem as a maximum a-posteriori (MAP) problem and develop a framework consisting of two levels of graphs to address it. The first-level graph, constructed as a weighted graph, is designed to efficiently and optimally infer initial values of scan poses from the unordered set. The second-level graph is constructed as a factor graph. By globally optimizing the variables on the graph, including scan poses, marker poses, and marker corner positions, we tackle the MAP problem. We conduct both qualitative and quantitative experiments to demonstrate that the proposed method surpasses previous state-of-the-art (SOTA) methods and to showcase that L-PR can serve as a low-cost and efficient tool for 3D asset collection and training data collection. In particular, we collect a new dataset named Livox-3DMatch using L-PR and incorporate it into the training of the SOTA learning-based method, SGHR, which brings evident improvements for SGHR on various benchmarks.

Tianpei Liao, Amal Haridevan, Yibo Liu et al.

There is a risk of collision when multiple UAVs land simultaneously without communication on the same platform. This work accomplishes vision-based autonomous landing and uses a deep-learning-based method to realize collision avoidance during the landing process.

Yibo Liu, Hunter Schofield, Jinjun Shan

Navigation using only one marker, which contains four artificial features, is a challenging task since camera pose estimation using only four coplanar points suffers from the rotational ambiguity problem in a real-world application. This paper presents a framework of vision-based navigation for a self-driving vehicle equipped with multiple cameras and a wheel odometer. A multiple camera setup is presented for the camera cluster which has 360-degree vision such that our framework solely requires one planar marker. A Kalman-Filter-based fusion method is introduced for the multiple-camera and wheel odometry. Furthermore, an algorithm is proposed to resolve the rotational ambiguity problem using the prediction of the Kalman Filter as additional information. Finally, the lateral and longitudinal controllers are provided. Experiments are conducted to illustrate the effectiveness of the theory.

Mengyu Zhou, Qingtao Li, Xinyi He et al.

It is common for people to create different types of charts to explore a multi-dimensional dataset (table). However, to recommend commonly composed charts in real world, one should take the challenges of efficiency, imbalanced data and table context into consideration. In this paper, we propose Table2Charts framework which learns common patterns from a large corpus of (table, charts) pairs. Based on deep Q-learning with copying mechanism and heuristic searching, Table2Charts does table-to-sequence generation, where each sequence follows a chart template. On a large spreadsheet corpus with 165k tables and 266k charts, we show that Table2Charts could learn a shared representation of table fields so that recommendation tasks on different chart types could mutually enhance each other. Table2Charts outperforms other chart recommendation systems in both multi-type task (with doubled recall numbers R@3=0.61 and R@1=0.43) and human evaluations.

Zhaoyi Wan, Fengming Xie, Yibo Liu et al.

Scene text recognition has been an important, active research topic in computer vision for years. Previous approaches mainly consider text as 1D signals and cast scene text recognition as a sequence prediction problem, by feat of CTC or attention based encoder-decoder framework, which is originally designed for speech recognition. However, different from speech voices, which are 1D signals, text instances are essentially distributed in 2D image spaces. To adhere to and make use of the 2D nature of text for higher recognition accuracy, we extend the vanilla CTC model to a second dimension, thus creating 2D-CTC. 2D-CTC can adaptively concentrate on most relevant features while excluding the impact from clutters and noises in the background; It can also naturally handle text instances with various forms (horizontal, oriented and curved) while giving more interpretable intermediate predictions. The experiments on standard benchmarks for scene text recognition, such as IIIT-5K, ICDAR 2015, SVP-Perspective, and CUTE80, demonstrate that the proposed 2D-CTC model outperforms state-of-the-art methods on the text of both regular and irregular shapes. Moreover, 2D-CTC exhibits its superiority over prior art on training and testing speed. Our implementation and models of 2D-CTC will be made publicly available soon later.