Hang Zhang

Zhenghao Lin, Yeyun Gong, Xiao Liu et al. · microsoft-research

Knowledge distillation is an effective way to transfer knowledge from a strong teacher to an efficient student model. Ideally, we expect the better the teacher is, the better the student. However, this expectation does not always come true. It is common that a better teacher model results in a bad student via distillation due to the nonnegligible gap between teacher and student. To bridge the gap, we propose PROD, a PROgressive Distillation method, for dense retrieval. PROD consists of a teacher progressive distillation and a data progressive distillation to gradually improve the student. We conduct extensive experiments on five widely-used benchmarks, MS MARCO Passage, TREC Passage 19, TREC Document 19, MS MARCO Document and Natural Questions, where PROD achieves the state-of-the-art within the distillation methods for dense retrieval. The code and models will be released.

Shuai Fan, Chen Lin, Haonan Li et al.

Most existing pre-trained language representation models (PLMs) are sub-optimal in sentiment analysis tasks, as they capture the sentiment information from word-level while under-considering sentence-level information. In this paper, we propose SentiWSP, a novel Sentiment-aware pre-trained language model with combined Word-level and Sentence-level Pre-training tasks. The word level pre-training task detects replaced sentiment words, via a generator-discriminator framework, to enhance the PLM's knowledge about sentiment words. The sentence level pre-training task further strengthens the discriminator via a contrastive learning framework, with similar sentences as negative samples, to encode sentiments in a sentence. Extensive experimental results show that SentiWSP achieves new state-of-the-art performance on various sentence-level and aspect-level sentiment classification benchmarks. We have made our code and model publicly available at https://github.com/XMUDM/SentiWSP.

Jinwei Zhang, Pascal Spincemaille, Hang Zhang et al.

Quantitative susceptibility mapping (QSM) involves acquisition and reconstruction of a series of images at multi-echo time points to estimate tissue field, which prolongs scan time and requires specific reconstruction technique. In this paper, we present our new framework, called Learned Acquisition and Reconstruction Optimization (LARO), which aims to accelerate the multi-echo gradient echo (mGRE) pulse sequence for QSM. Our approach involves optimizing a Cartesian multi-echo k-space sampling pattern with a deep reconstruction network. Next, this optimized sampling pattern was implemented in an mGRE sequence using Cartesian fan-beam k-space segmenting and ordering for prospective scans. Furthermore, we propose to insert a recurrent temporal feature fusion module into the reconstruction network to capture signal redundancies along echo time. Our ablation studies show that both the optimized sampling pattern and proposed reconstruction strategy help improve the quality of the multi-echo image reconstructions. Generalization experiments show that LARO is robust on the test data with new pathologies and different sequence parameters. Our code is available at https://github.com/Jinwei1209/LARO.git.

Hang Zhang, Xin Li, Lidong Bing

We present Video-LLaMA a multi-modal framework that empowers Large Language Models (LLMs) with the capability of understanding both visual and auditory content in the video. Video-LLaMA bootstraps cross-modal training from the frozen pre-trained visual and audio encoders and the frozen LLMs. Unlike previous works that complement LLMs to process the visual or audio signals only, Video-LLaMA enables video comprehension by tackling two challenges: (1) capturing the temporal changes in visual scenes, (2) integrating audio-visual signals. To counter the first challenge, we propose a Video Q-former to assemble a pre-trained image encoder into our video encoder and introduce a video-to-text generation task to learn video-language correspondence. For the second challenge, we leverage ImageBind, a universal embedding model aligning multiple modalities, as the pre-trained audio encoder and introduce an Audio Q-former on top of ImageBind to learn reasonable auditory query embeddings for the LLM module. To align the output of both visual and audio encoders with LLM's embedding space, we first train Video-LLaMA on massive video/image-caption pairs and then tune our model with visual-instruction datasets of moderate amount but higher quality. We found Video-LLaMA shows the ability to perceive and comprehend video content and generate meaningful responses grounded in the visual and auditory information presented in the videos.

Xingwei He, Zhenghao Lin, Yeyun Gong et al.

Many natural language processing (NLP) tasks rely on labeled data to train machine learning models with high performance. However, data annotation is time-consuming and expensive, especially when the task involves a large amount of data or requires specialized domains. Recently, GPT-3.5 series models have demonstrated remarkable few-shot and zero-shot ability across various NLP tasks. In this paper, we first claim that large language models (LLMs), such as GPT-3.5, can serve as an excellent crowdsourced annotator when provided with sufficient guidance and demonstrated examples. Accordingly, we propose AnnoLLM, an annotation system powered by LLMs, which adopts a two-step approach, explain-then-annotate. Concretely, we first prompt LLMs to provide explanations for why the specific ground truth answer/label was assigned for a given example. Then, we construct the few-shot chain-of-thought prompt with the self-generated explanation and employ it to annotate the unlabeled data with LLMs. Our experiment results on three tasks, including user input and keyword relevance assessment, BoolQ, and WiC, demonstrate that AnnoLLM surpasses or performs on par with crowdsourced annotators. Furthermore, we build the first conversation-based information retrieval dataset employing AnnoLLM. This dataset is designed to facilitate the development of retrieval models capable of retrieving pertinent documents for conversational text. Human evaluation has validated the dataset's high quality.

Feng Liang, Bichen Wu, Xiaoliang Dai et al.

Open-vocabulary semantic segmentation aims to segment an image into semantic regions according to text descriptions, which may not have been seen during training. Recent two-stage methods first generate class-agnostic mask proposals and then leverage pre-trained vision-language models, e.g., CLIP, to classify masked regions. We identify the performance bottleneck of this paradigm to be the pre-trained CLIP model, since it does not perform well on masked images. To address this, we propose to finetune CLIP on a collection of masked image regions and their corresponding text descriptions. We collect training data by mining an existing image-caption dataset (e.g., COCO Captions), using CLIP to match masked image regions to nouns in the image captions. Compared with the more precise and manually annotated segmentation labels with fixed classes (e.g., COCO-Stuff), we find our noisy but diverse dataset can better retain CLIP's generalization ability. Along with finetuning the entire model, we utilize the "blank" areas in masked images using a method we dub mask prompt tuning. Experiments demonstrate mask prompt tuning brings significant improvement without modifying any weights of CLIP, and it can further improve a fully finetuned model. In particular, when trained on COCO and evaluated on ADE20K-150, our best model achieves 29.6% mIoU, which is +8.5% higher than the previous state-of-the-art. For the first time, open-vocabulary generalist models match the performance of supervised specialist models in 2017 without dataset-specific adaptations.

Sicong Leng, Hang Zhang, Guanzheng Chen et al.

Large Vision-Language Models (LVLMs) have advanced considerably, intertwining visual recognition and language understanding to generate content that is not only coherent but also contextually attuned. Despite their success, LVLMs still suffer from the issue of object hallucinations, where models generate plausible yet incorrect outputs that include objects that do not exist in the images. To mitigate this issue, we introduce Visual Contrastive Decoding (VCD), a simple and training-free method that contrasts output distributions derived from original and distorted visual inputs. The proposed VCD effectively reduces the over-reliance on statistical bias and unimodal priors, two essential causes of object hallucinations. This adjustment ensures the generated content is closely grounded to visual inputs, resulting in contextually accurate outputs. Our experiments show that VCD, without either additional training or the usage of external tools, significantly mitigates the object hallucination issue across different LVLM families. Beyond mitigating object hallucinations, VCD also excels in general LVLM benchmarks, highlighting its wide-ranging applicability.

Hang Zhang, Rongguang Wang, Renjiu Hu et al.

Chronic active multiple sclerosis lesions, also termed as rim+ lesions, can be characterized by a hyperintense rim at the edge of the lesion on quantitative susceptibility maps. These rim+ lesions exhibit a geometrically simple structure, where gradients at the lesion edge are radially oriented and a greater magnitude of gradients is observed in contrast to rim- (non rim+) lesions. However, recent studies have shown that the identification performance of such lesions remains unsatisfied due to the limited amount of data and high class imbalance. In this paper, we propose a simple yet effective image processing operation, deep directed accumulator (DeDA), that provides a new perspective for injecting domain-specific inductive biases (priors) into neural networks for rim+ lesion identification. Given a feature map and a set of sampling grids, DeDA creates and quantizes an accumulator space into finite intervals, and accumulates feature values accordingly. This DeDA operation is a generalized discrete Radon transform and can also be regarded as a symmetric operation to the grid sampling within the forward-backward neural network framework, the process of which is order-agnostic, and can be efficiently implemented with the native CUDA programming. Experimental results on a dataset with 177 rim+ and 3986 rim- lesions show that 10.1% of improvement in a partial (false positive rate<0.1) area under the receiver operating characteristic curve (pROC AUC) and 10.2% of improvement in an area under the precision recall curve (PR AUC) can be achieved respectively comparing to other state-of-the-art methods. The source code is available online at https://github.com/tinymilky/DeDA

Hang Zhang, Renjiu Hu, Xiang Chen et al.

Recent research highlights that the Directed Accumulator (DA), through its parametrization of geometric priors into neural networks, has notably improved the performance of medical image recognition, particularly with small and imbalanced datasets. However, DA's potential in pixel-wise dense predictions is unexplored. To bridge this gap, we present the Directed Accumulator Grid (DAGrid), which allows geometric-preserving filtering in neural networks, thus broadening the scope of DA's applications to include pixel-level dense prediction tasks. DAGrid utilizes homogeneous data types in conjunction with designed sampling grids to construct geometrically transformed representations, retaining intricate geometric information and promoting long-range information propagation within the neural networks. Contrary to its symmetric counterpart, grid sampling, which might lose information in the sampling process, DAGrid aggregates all pixels, ensuring a comprehensive representation in the transformed space. The parallelization of DAGrid on modern GPUs is facilitated using CUDA programming, and also back propagation is enabled for deep neural network training. Empirical results show DAGrid-enhanced neural networks excel in supervised skin lesion segmentation and unsupervised cardiac image registration. Specifically, the network incorporating DAGrid has realized a 70.8% reduction in network parameter size and a 96.8% decrease in FLOPs, while concurrently improving the Dice score for skin lesion segmentation by 1.0% compared to state-of-the-art transformers. Furthermore, it has achieved improvements of 4.4% and 8.2% in the average Dice score and Dice score of the left ventricular mass, respectively, indicating an increase in registration accuracy for cardiac images. The source code is available at https://github.com/tinymilky/DeDA.

Xingwei He, Yeyun Gong, A-Long Jin et al.

Commonsense generation aims to generate a realistic sentence describing a daily scene under the given concepts, which is very challenging, since it requires models to have relational reasoning and compositional generalization capabilities. Previous work focuses on retrieving prototype sentences for the provided concepts to assist generation. They first use a sparse retriever to retrieve candidate sentences, then re-rank the candidates with a ranker. However, the candidates returned by their ranker may not be the most relevant sentences, since the ranker treats all candidates equally without considering their relevance to the reference sentences of the given concepts. Another problem is that re-ranking is very expensive, but only using retrievers will seriously degrade the performance of their generation models. To solve these problems, we propose the metric distillation rule to distill knowledge from the metric (e.g., BLEU) to the ranker. We further transfer the critical knowledge summarized by the distilled ranker to the retriever. In this way, the relevance scores of candidate sentences predicted by the ranker and retriever will be more consistent with their quality measured by the metric. Experimental results on the CommonGen benchmark verify the effectiveness of our proposed method: (1) Our generation model with the distilled ranker achieves a new state-of-the-art result. (2) Our generation model with the distilled retriever even surpasses the previous SOTA.

Jiashuo Sun, Hang Zhang, Chen Lin et al.

Long-form numerical reasoning in financial analysis aims to generate a reasoning program to calculate the correct answer for a given question. Previous work followed a retriever-generator framework, where the retriever selects key facts from a long-form document, and the generator generates a reasoning program based on retrieved facts. However, they treated all facts equally without considering the different contributions of facts with and without numbers. Meanwhile, the program consistency were ignored under supervised training, resulting in lower training accuracy and diversity. To solve these problems, we proposed APOLLO to improve the long-form numerical reasoning framework. For the retriever, we adopt a number-aware negative sampling strategy to enable the retriever to be more discriminative on key numerical facts. For the generator, we design consistency-based reinforcement learning and target program augmentation strategy based on the consistency of program execution results. Experimental results on the FinQA and ConvFinQA leaderboard verify the effectiveness of our proposed method, achieving the new state-of-the-art.

Xingwei He, Yeyun Gong, A-Long Jin et al.

The dual-encoder has become the de facto architecture for dense retrieval. Typically, it computes the latent representations of the query and document independently, thus failing to fully capture the interactions between the query and document. To alleviate this, recent research has focused on obtaining query-informed document representations. During training, it expands the document with a real query, but during inference, it replaces the real query with a generated one. This inconsistency between training and inference causes the dense retrieval model to prioritize query information while disregarding the document when computing the document representation. Consequently, it performs even worse than the vanilla dense retrieval model because its performance heavily relies on the relevance between the generated queries and the real query.In this paper, we propose a curriculum sampling strategy that utilizes pseudo queries during training and progressively enhances the relevance between the generated query and the real query. By doing so, the retrieval model learns to extend its attention from the document alone to both the document and query, resulting in high-quality query-informed document representations. Experimental results on both in-domain and out-of-domain datasets demonstrate that our approach outperforms previous dense retrieval models.

Hang Zhang, Xiang Chen, Renjiu Hu et al.

Many existing learning-based deformable image registration methods impose constraints on deformation fields to ensure they are globally smooth and continuous. However, this assumption does not hold in cardiac image registration, where different anatomical regions exhibit asymmetric motions during respiration and movements due to sliding organs within the chest. Consequently, such global constraints fail to accommodate local discontinuities across organ boundaries, potentially resulting in erroneous and unrealistic displacement fields. In this paper, we address this issue with MemWarp, a learning framework that leverages a memory network to store prototypical information tailored to different anatomical regions. MemWarp is different from earlier approaches in two main aspects: firstly, by decoupling feature extraction from similarity matching in moving and fixed images, it facilitates more effective utilization of feature maps; secondly, despite its capability to preserve discontinuities, it eliminates the need for segmentation masks during model inference. In experiments on a publicly available cardiac dataset, our method achieves considerable improvements in registration accuracy and producing realistic deformations, outperforming state-of-the-art methods with a remarkable 7.1\% Dice score improvement over the runner-up semi-supervised method. Source code will be available at https://github.com/tinymilky/Mem-Warp.

Muralikrishnna G. Sethuraman, Hang Zhang, Faramarz Fekri · gatech

In this paper, we propose a general framework for designing sensing matrix $\boldsymbol{A} \in \mathbb{R}^{d\times p}$, for estimation of sparse covariance matrix from compressed measurements of the form $\boldsymbol{y} = \boldsymbol{A}\boldsymbol{x} + \boldsymbol{n}$, where $\boldsymbol{y}, \boldsymbol{n} \in \mathbb{R}^d$, and $\boldsymbol{x} \in \mathbb{R}^p$. By viewing covariance recovery as inference over factor graphs via message passing algorithm, ideas from coding theory, such as \textit{Density Evolution} (DE), are leveraged to construct a framework for the design of the sensing matrix. The proposed framework can handle both (1) regular sensing, i.e., equal importance is given to all entries of the covariance, and (2) preferential sensing, i.e., higher importance is given to a part of the covariance matrix. Through experiments, we show that the sensing matrix designed via density evolution can match the state-of-the-art for covariance recovery in the regular sensing paradigm and attain improved performance in the preferential sensing regime. Additionally, we study the feasibility of causal graph structure recovery using the estimated covariance matrix obtained from the compressed measurements.

Lingjun Zhang, Yujian Yuan, Changjie Wu et al.

Vision-Language Models (VLM) exhibit strong reasoning capabilities, showing promise for end-to-end autonomous driving systems. Chain-of-Thought (CoT), as VLM's widely used reasoning strategy, is facing critical challenges. Existing textual CoT has a large gap between text semantic space and trajectory physical space. Although the recent approach utilizes future image to replace text as CoT process, it lacks clear planning-oriented objective guidance to generate images with accurate scene evolution. To address these, we innovatively propose MindDriver, a progressive multimodal reasoning framework that enables VLM to imitate human-like progressive thinking for autonomous driving. MindDriver presents semantic understanding, semantic-to-physical space imagination, and physical-space trajectory planning. To achieve aligned reasoning processes in MindDriver, we develop a feedback-guided automatic data annotation pipeline to generate aligned multimodal reasoning training data. Furthermore, we develop a progressive reinforcement fine-tuning method to optimize the alignment through progressive high- level reward-based learning. MindDriver demonstrates superior performance in both nuScences open-loop and Bench2Drive closed-loop evaluation. Codes are available at https://github.com/hotdogcheesewhite/MindDriver.

Zehan Wang, Ziang Zhang, Hang Zhang et al.

Recently, human-computer interaction with various modalities has shown promising applications, like GPT-4o and Gemini. Given the foundational role of multimodal joint representation in understanding and generation pipelines, high-quality omni joint representations would be a step toward co-processing more diverse multimodal information. In this work, we present OmniBind, large-scale multimodal joint representation models ranging in scale from 7 billion to 30 billion parameters, which support 3D, audio, image, and language inputs. Due to the scarcity of data pairs across all modalities, instead of training large models from scratch, we propose remapping and binding the spaces of various pre-trained specialist models together. This approach enables "scaling up" by indirectly increasing the model parameters and the amount of seen data. To effectively integrate various spaces, we dynamically assign weights to different spaces by learning routers with two objectives: cross-modal overall alignment and language representation decoupling. Notably, since binding and routing spaces both only require lightweight networks, OmniBind is extremely training-efficient. Learning the largest 30B model requires merely unpaired unimodal data and approximately 3 days on a single 8-4090 node. Extensive experiments demonstrate the versatility and superiority of OmniBind as an omni representation model, highlighting its great potential for diverse applications, such as any-query and composable multimodal understanding.

Jiazheng Wang, Xiang Chen, Yuxi Zhang et al.

Surgical navigation based on multimodal image registration has played a significant role in providing intraoperative guidance to surgeons by showing the relative position of the target area to critical anatomical structures during surgery. However, due to the differences between multimodal images and intraoperative image deformation caused by tissue displacement and removal during surgery, effective registration of preoperative and intraoperative multimodal images faces significant challenges. To address the multimodal image registration challenges in Learn2Reg 2024, an unsupervised multimodal medical image registration method based on multilevel correlation balanced optimization (MCBO) is designed to solve these problems. First, the features of each modality are extracted based on the modality independent neighborhood descriptor, and the multimodal images are mapped to the feature space. Second, a multilevel pyramidal fusion optimization mechanism is designed to achieve global optimization and local detail complementation of the deformation field through dense correlation analysis and weight-balanced coupled convex optimization for input features at different scales. For preoperative medical images in different modalities, the alignment and stacking of valid information between different modalities is achieved by the maximum fusion between deformation fields. Our method focuses on the ReMIND2Reg task in Learn2Reg 2024, and to verify the generality of the method, we also tested it on the COMULIS3DCLEM task. Based on the results, our method achieved second place in the validation of both two tasks. The code is available at https://github.com/wjiazheng/MCBO.

Yuting Xu, Jiayi Tian, Jian Liang et al.

Existing benchmarks have laid the foundation for travel planning agents by establishing API-centric paradigms. However, as the capabilities of Autonomous Agents continue to advance, their evaluation must evolve beyond simple tool execution toward handling the inherent complexities of the open web. Current benchmarks bypass core cognitive hurdles: they fail to account for information noise, ignore multi-source factual contradictions, and overlook the necessity of grounding visual perception into logical planning. We introduce VeriTrip, a verifiable benchmark designed to meet the increasing demands for agent robustness and reliability. VeriTrip shifts the evaluation focus to evidence-grounded reasoning over unstructured multimodal web corpora. It establishes a Multimodal Retrieval Base (MRB) derived from real-world sources, forcing agents to autonomously orchestrate queries across heterogeneous data. A synchronized Verifiable Knowledge Base (VKB) enables a cell-wise verification protocol that precisely quantifies factual reliability, distinguishing systematic reasoning failures from parametric hallucinations. Our evaluations across leading MLLMs reveal a critical \textit{retrieval-reasoning trade-off}: the cognitive load of autonomous retrieval significantly erodes instruction retention. VeriTrip provides the rigorous foundation necessary for the next generation of planning agents capable of operating in unconstrained, multimodal environments.

Hang Zhang, Yeyun Gong, Xingwei He et al.

Most dense retrieval models contain an implicit assumption: the training query-document pairs are exactly matched. Since it is expensive to annotate the corpus manually, training pairs in real-world applications are usually collected automatically, which inevitably introduces mismatched-pair noise. In this paper, we explore an interesting and challenging problem in dense retrieval, how to train an effective model with mismatched-pair noise. To solve this problem, we propose a novel approach called Noisy Pair Corrector (NPC), which consists of a detection module and a correction module. The detection module estimates noise pairs by calculating the perplexity between annotated positive and easy negative documents. The correction module utilizes an exponential moving average (EMA) model to provide a soft supervised signal, aiding in mitigating the effects of noise. We conduct experiments on text-retrieval benchmarks Natural Question and TriviaQA, code-search benchmarks StaQC and SO-DS. Experimental results show that NPC achieves excellent performance in handling both synthetic and realistic noise.

Jiaxu Wan, Xu Wang, Mengwei Xie et al.

Recent spatial intelligence approaches typically attach 3D cues to 2D reasoning pipelines or couple MLLMs with black-box reconstruction modules, leading to weak spatial consistency, limited viewpoint diversity, and evidence chains that cannot be traced back to supporting views. Frameworks for "thinking with images" (e.g., ChatGPT-o3 and DeepEyes) show that stepwise multimodal reasoning can emerge by interleaving hypothesis formation with active acquisition of visual evidence, but they do not address three key challenges in spatial Chain-of-Thought (CoT): building global space perception under strict token budgets, explicitly associating 3D hypotheses with video frames for verification, and designing spatially grounded rewards for reinforcement learning. To address these issues, we present EagleVision, a dual-stage framework for progressive spatial cognition through macro perception and micro verification. In the macro perception stage, EagleVision employs a semantics-perspective-fusion determinantal point process (SPF-DPP) to select a compact set of geometry- and semantics-aware keyframes from long videos under a fixed token budget. In the micro verification stage, we formalize spatial CoT as BEV-grounded pose querying: the agent iteratively predicts poses on a BEV plane, retrieves the nearest real frames, and is trained purely by reinforcement learning with a spatial grounding reward that scores the consistency between predicted poses and observed views. On VSI-Bench, EagleVision achieves state-of-the-art performance among open-source vision-language models, demonstrating strong and generalizable spatial understanding.

Xiang Chen, Min Liu, Rongguang Wang et al.

Medical images are often characterized by their structured anatomical representations and spatially inhomogeneous contrasts. Leveraging anatomical priors in neural networks can greatly enhance their utility in resource-constrained clinical settings. Prior research has harnessed such information for image segmentation, yet progress in deformable image registration has been modest. Our work introduces textSCF, a novel method that integrates spatially covariant filters and textual anatomical prompts encoded by visual-language models, to fill this gap. This approach optimizes an implicit function that correlates text embeddings of anatomical regions to filter weights, relaxing the typical translation-invariance constraint of convolutional operations. TextSCF not only boosts computational efficiency but can also retain or improve registration accuracy. By capturing the contextual interplay between anatomical regions, it offers impressive inter-regional transferability and the ability to preserve structural discontinuities during registration. TextSCF's performance has been rigorously tested on inter-subject brain MRI and abdominal CT registration tasks, outperforming existing state-of-the-art models in the MICCAI Learn2Reg 2021 challenge and leading the leaderboard. In abdominal registrations, textSCF's larger model variant improved the Dice score by 11.3% over the second-best model, while its smaller variant maintained similar accuracy but with an 89.13% reduction in network parameters and a 98.34\% decrease in computational operations.

Ming Qian, Zimin Xia, Changkun Liu et al.

Generating a street-level 3D scene from a single satellite image is a crucial yet challenging task. Current methods present a stark trade-off: geometry-colorization models achieve high geometric fidelity but are typically building-focused and lack semantic diversity. In contrast, proxy-based models use feed-forward image-to-3D frameworks to generate holistic scenes by jointly learning geometry and texture, a process that yields rich content but coarse and unstable geometry. We attribute these geometric failures to the extreme viewpoint gap and sparse, inconsistent supervision inherent in satellite-to-street data. We introduce Sat3DGen to address these fundamental challenges, which embodies a geometry-first methodology. This methodology enhances the feed-forward paradigm by integrating novel geometric constraints with a perspective-view training strategy, explicitly countering the primary sources of geometric error. This geometry-centric strategy yields a dramatic leap in both 3D accuracy and photorealism. For validation, we first constructed a new benchmark by pairing the VIGOR-OOD test set with high-resolution DSM data. On this benchmark, our method improves geometric RMSE from 6.76m to 5.20m. Crucially, this geometric leap also boosts photorealism, reducing the Fréchet Inception Distance (FID) from $\sim$40 to 19 against the leading method, Sat2Density++, despite using no extra tailored image-quality modules. We demonstrate the versatility of our high-quality 3D assets through diverse downstream applications, including semantic-map-to-3D synthesis, multi-camera video generation, large-scale meshing, and unsupervised single-image Digital Surface Model (DSM) estimation. The code has been released on https://github.com/qianmingduowan/Sat3DGen.

Jiazheng Wang, Zeyu Liu, Min Liu et al.

Surgical navigation based on multimodal image registration has played a significant role in providing intraoperative guidance to surgeons by showing the relative position of the target area to critical anatomical structures during surgery. However, due to the differences between multimodal images and intraoperative image deformation caused by tissue displacement and removal during the surgery, effective registration of preoperative and intraoperative multimodal images faces significant challenges. To address the multimodal image registration challenges in Learn2Reg 2025, an unsupervised multimodal medical image registration method based on Multilevel Correlation Pyramidal Optimization (MCPO) is designed to solve these problems. First, the features of each modality are extracted based on the modality independent neighborhood descriptor, and the multimodal images is mapped to the feature space. Second, a multilevel pyramidal fusion optimization mechanism is designed to achieve global optimization and local detail complementation of the displacement field through dense correlation analysis and weight-balanced coupled convex optimization for input features at different scales. Our method focuses on the ReMIND2Reg task in Learn2Reg 2025. Based on the results, our method achieved the first place in the validation phase and test phase of ReMIND2Reg. The MCPO is also validated on the Resect dataset, achieving an average TRE of 1.798 mm. This demonstrates the broad applicability of our method in preoperative-to-intraoperative image registration. The code is available at https://github.com/wjiazheng/MCPO.

Yuyang Tan, Renhe Zhang, Hang Zhang et al.

Referring 3D Gaussian Splatting (R3DGS), which utilizes natural language for 3D object segmentation, has emerged as a crucial capability for embodied AI. However, existing methods typically rely on expensive per-scene manual annotation and per-view pseudo mask generation, which suffer from multi-view inconsistency and poor generalization to varying query specificities. To address this, we present TrackRef3D, a fully automatic pipeline that achieves open-world referring segmentation in 3D Gaussian Splatting (3DGS) without manual annotation by introducing a multi-view consistent track-then-label paradigm that fundamentally decouples object discovery from semantic grounding. Specifically, we propose a Trajectory-Aware Semantic Consensus Module (TSCM) which aggregates cross-view predictions via synonymous clustering and trajectory-aware voting to establish a canonical semantic identity, thereby ensuring multi-view consistency. Furthermore, we employ a visibility-aware description generation strategy to mitigate ambiguity and propose a Hybrid Training Strategy (HTS) that jointly optimizes coarse category semantics and fine-grained referential cues to ensure robustness under varying query specificities using a multi-positive contrastive objective. Extensive experiments on benchmarks demonstrate that TrackRef3D achieves state-of-the-art performance.

Jiangyang Li, Cong Wan, Changjie Wu et al.

Reliable spatial reasoning remains a core bottleneck for vision-language models (VLMs). Existing mainstream training paradigms for spatial reasoning largely rely on outcome alignment or process imitation, lacking explicit constraints on the reasoning process, and therefore struggle to ensure genuine visual dependence and stable reasoning trajectories. In this paper, we construct a high-quality CoT dataset covering diverse spatial phenomena and diagnose the model's reasoning process, revealing two typical types of process degradation during reinforcement learning optimization: Spurious Grounding, which bypasses visual evidence, and Tail Instability, where uncertainty abnormally rises in the later stage of reasoning. To address these issues, we propose ProSR, a process-shaping optimization framework for spatial reasoning. Through a Counterfactual Invariance Penalty and a Tail Drift Penalty, ProSR extends the optimization objective from single answer correctness to two process-level dimensions: visual dependence and trajectory stability. Experiments on multiple complex and out-of-distribution spatial reasoning benchmarks show that ProSR improves answer accuracy while generating reasoning trajectories that are more stable and more dependent on visual evidence.

Yuelyu Ji, Min Gu Kwak, Hang Zhang et al.

Medical RAG needs evidence-grounded claims, so plugging a claim-level NLI checker into retrieval-augmented RL is intuitive. \textbf{We find that the checker's \emph{output distribution} during training, not its held-out accuracy, decides whether it provides trainable gradient.} We compare four NLI checker back-ends as process rewards inside a GRPO-trained medical RAG agent (Qwen2.5-7B, replicated on Qwen3-4B and Llama-3.1-8B) across four held-out medical QA benchmarks. Three diagnostic findings emerge. \textbf{(i)} Signal collapse is log-prob-specific: LLM log-probability scoring labels over 97\% of claims neutral -- collapsing the RL gradient to zero -- while a calibrated MedNLI classifier scores the same pairs non-degenerately. \textbf{(ii)} Moderate signal beats strong signal on answer quality: a strong proprietary checker triggers a three-step reward-hacking cascade -- ultra-short answers, search avoidance, language collapse -- so a moderate-signal local classifier trains a higher-quality model (\textbf{+12\% BERTScore over zero-shot, no GPT dependency}). \textbf{(iii)} Signal strength is policy-dependent: the same checker registers as moderate on one policy but strong on another without triggering the cascade end-state. We frame these as boundary conditions for verifier-as-reward systems.

Hang Zhang, Rongguang Wang, Jinwei Zhang et al.

Compared to natural images, medical images usually show stronger visual patterns and therefore this adds flexibility and elasticity to resource-limited clinical applications by injecting proper priors into neural networks. In this paper, we propose spatially covariant pixel-aligned classifier (SCP) to improve the computational efficiency and meantime maintain or increase accuracy for lesion segmentation. SCP relaxes the spatial invariance constraint imposed by convolutional operations and optimizes an underlying implicit function that maps image coordinates to network weights, the parameters of which are obtained along with the backbone network training and later used for generating network weights to capture spatially covariant contextual information. We demonstrate the effectiveness and efficiency of the proposed SCP using two lesion segmentation tasks from different imaging modalities: white matter hyperintensity segmentation in magnetic resonance imaging and liver tumor segmentation in contrast-enhanced abdominal computerized tomography. The network using SCP has achieved 23.8%, 64.9% and 74.7% reduction in GPU memory usage, FLOPs, and network size with similar or better accuracy for lesion segmentation.

Lingjun Zhang, Changjie Wu, Linzhe Shi et al.

End-to-end autonomous driving systems are increasingly integrating Vision-Language Model (VLM) architectures, incorporating text reasoning or visual reasoning to enhance the robustness and accuracy of driving decisions. However, the reasoning mechanisms employed in most methods are direct adaptations from general domains, lacking in-depth exploration tailored to autonomous driving scenarios, particularly within visual reasoning modules. In this paper, we propose a driving world model that performs parallel prediction of latent semantic features for consecutive future frames in the bird's-eye-view (BEV) space, thereby enabling long-horizon modeling of future world states. We also introduce an efficient and adaptive text reasoning mechanism that utilizes additional social knowledge and reasoning capabilities to further improve driving performance in challenging long-tail scenarios. We present a novel, efficient, and effective approach that achieves state-of-the-art (SOTA) results on the closed-loop Bench2drive benchmark. Codes are available at: https://github.com/hotdogcheesewhite/DeepSight.

Xizhi Wu, Nelly Estefanie Garduno-Rapp, Justin F Rousseau et al.

Unlike most primary headaches, secondary headaches need specialized care and can have devastating consequences if not treated promptly. Clinical guidelines highlight several 'red flag' features, such as thunderclap onset, meningismus, papilledema, focal neurologic deficits, signs of temporal arteritis, systemic illness, and the 'worst headache of their life' presentation. Despite these guidelines, determining which patients require urgent evaluation remains challenging in primary care settings. Clinicians often work with limited time, incomplete information, and diverse symptom presentations, which can lead to under-recognition and inappropriate care. We present a large language model (LLM)-based multi-agent clinical decision support system built on an orchestrator-specialist architecture, designed to perform explicit and interpretable secondary headache diagnosis from free-text clinical vignettes. The multi-agent system decomposes diagnosis into seven domain-specialized agents, each producing a structured and evidence-grounded rationale, while a central orchestrator performs task decomposition and coordinates agent routing. We evaluated the multi-agent system using 90 expert-validated secondary headache cases and compared its performance with a single-LLM baseline across two prompting strategies: question-based prompting (QPrompt) and clinical practice guideline-based prompting (GPrompt). We tested five open-source LLMs (Qwen-30B, GPT-OSS-20B, Qwen-14B, Qwen-8B, and Llama-3.1-8B), and found that the orchestrated multi-agent system with GPrompt consistently achieved the highest F1 scores, with larger gains in smaller models. These findings demonstrate that structured multi-agent reasoning improves accuracy beyond prompt engineering alone and offers a transparent, clinically aligned approach for explainable decision support in secondary headache diagnosis.

Hang Zhang, Afshin Abdi, Faramarz Fekri

This paper considers learning of the graphical structure of a $p$-dimensional random vector $X \in R^p$ using both parametric and non-parametric methods. Unlike the previous works which observe $x$ directly, we consider the indirect observation scenario in which samples $y$ are collected via a sensing matrix $A \in R^{d\times p}$, and corrupted with some additive noise $w$, i.e, $Y = AX + W$. For the parametric method, we assume $X$ to be Gaussian, i.e., $x\in R^p\sim N(μ, Σ)$ and $Σ\in R^{p\times p}$. For the first time, we show that the correct graphical structure can be correctly recovered under the indefinite sensing system ($d < p$) using insufficient samples ($n < p$). In particular, we show that for the exact recovery, we require dimension $d = Ω(p^{0.8})$ and sample number $n = Ω(p^{0.8}\log^3 p)$. For the nonparametric method, we assume a nonparanormal distribution for $X$ rather than Gaussian. Under mild conditions, we show that our graph-structure estimator can obtain the correct structure. We derive the minimum sample number $n$ and dimension $d$ as $n\gtrsim (deg)^4 \log^4 n$ and $d \gtrsim p + (deg\cdot\log(d-p))^{β/4}$, respectively, where deg is the maximum Markov blanket in the graphical model and $β> 0$ is some fixed positive constant. Additionally, we obtain a non-asymptotic uniform bound on the estimation error of the CDF of $X$ from indirect observations with inexact knowledge of the noise distribution. To the best of our knowledge, this bound is derived for the first time and may serve as an independent interest. Numerical experiments on both real-world and synthetic data are provided confirm the theoretical results.

Yiyi Tao, Zhuoyue Wang, Hang Zhang et al.

The success of Vision Language Models (VLMs) on various vision-language tasks heavily relies on pre-training with large scale web-crawled datasets. However, the noisy and incomplete nature of web data makes dataset scale crucial for performance, rendering end-to-end training increasingly prohibitive. In this paper, we propose NEVLP, a noise-robust framework for efficient vision-language pre-training that requires less pre-training data. Specifically, we bridge the modality gap between a frozen image encoder and a large language model with a transformer and introduce two innovative learning strategies: noise-adaptive learning and concept-enhanced learning to mitigate the impact of noise. In noise-adaptive learning, we estimate the noise probability of each image-text pair based on the transformer's memorization effect and employ noise-adaptive regularization on image-text contrastive learning to condition cross-modal alignment. In concept-enhanced learning, we enrich incomplete text by incorporating visual concepts (objects in the image) to provide prior information about existing objects for image-text matching and image-grounded text generation, thereby mitigating text incompletion. Our framework effectively utilizes noisy web data and achieves state-of-the-art performance with less pre-training data across a wide range of vision-language tasks, including image-text retrieval, image captioning, and visual question answering.

Manas Gupta, Sarthak Ketanbhai Modi, Hang Zhang et al.

Bio-inspired learning has been gaining popularity recently given that Backpropagation (BP) is not considered biologically plausible. Many algorithms have been proposed in the literature which are all more biologically plausible than BP. However, apart from overcoming the biological implausibility of BP, a strong motivation for using Bio-inspired algorithms remains lacking. In this study, we undertake a holistic comparison of BP vs. multiple Bio-inspired algorithms to answer the question of whether Bio-learning offers additional benefits over BP. We test Bio-algorithms under different design choices such as access to only partial training data, resource constraints in terms of the number of training epochs, sparsification of the neural network parameters and addition of noise to input samples. Through these experiments, we notably find two key advantages of Bio-algorithms over BP. Firstly, Bio-algorithms perform much better than BP when the entire training dataset is not supplied. Four of the five Bio-algorithms tested outperform BP by upto 5% accuracy when only 20% of the training dataset is available. Secondly, even when the full dataset is available, Bio-algorithms learn much quicker and converge to a stable accuracy in far lesser training epochs than BP. Hebbian learning, specifically, is able to learn in just 5 epochs compared to around 100 epochs required by BP. These insights present practical reasons for utilising Bio-learning beyond just their biological plausibility and also point towards interesting new directions for future work on Bio-learning.

Tianhao Mao, Dongfang Zhao, Haixu Tang et al.

Large language models (LLMs) are rapidly transforming software engineering by enabling developers to generate code ranging from small snippets to entire projects. As AI-generated code becomes increasingly integrated into real-world systems, understanding its characteristics and impact is critical. However, prior work primarily focuses on small-scale, controlled evaluations and lacks comprehensive analysis in real-world settings. In this paper, we present a large-scale empirical study of AI-generated code in real-world repositories. We analyze both code-level metrics (\eg complexity, structure, and defect-related indicators) and commit-level characteristics (\eg commit size, frequency, and post-commit stability). To enable this study, we develop heuristic filter with LLM classification to identify AI-generated code and construct a large dataset. Our results provide new insights into how AI-generated code differs from human-written code and how AI assistance influences development practices. These findings contribute to a deeper understanding of the practical implications of AI-assisted programming.

Sifan Zhou, Hang Zhang, Yuhang Wang et al.

While 4D Gaussian Splatting (4DGS) has revolutionized high-fidelity dynamic reconstruction, safeguarding the intellectual property of these assets remains an open challenge. Conventional steganographic techniques often neglect the underlying kinematic manifolds, triggering non-physical artifacts such as severe temporal flickering and "FVD collapse". To address this, we propose \textbf{4D-GSW}, a kinematic-aware watermarking framework designed to embed robust copyright information while preserving high spatio-temporal consistency. Unlike prior 4D steganography that primarily focuses on opacity-guided invisibility, our approach explicitly addresses the physical coherence of motion trajectories. We introduce a \textbf{Spatio-Temporal Curvature (STC)} metric to identify "Dynamic Instants," adaptively gating watermark gradient injection to shield critical motion manifolds from non-physical perturbations. To ensure global coherence across complex deformations, we formulate a joint \textbf{HMM-MRF energy minimization} model that synchronizes watermark phases within both temporal trajectories and spatial neighborhoods. Furthermore, an \textbf{anisotropic gradient routing} mechanism ensures that watermark embedding remains strictly decoupled from photometric reconstruction fidelity. Extensive experiments have demonstrated the superior performance of our method in robustly hiding watermarks while resisting various attacks and maintaining high rendering quality and spatiotemporal consistency.

Yongxin Zhu, Bocheng Li, Hang Zhang et al.

Latent-based image generative models, such as Latent Diffusion Models (LDMs) and Mask Image Models (MIMs), have achieved notable success in image generation tasks. These models typically leverage reconstructive autoencoders like VQGAN or VAE to encode pixels into a more compact latent space and learn the data distribution in the latent space instead of directly from pixels. However, this practice raises a pertinent question: Is it truly the optimal choice? In response, we begin with an intriguing observation: despite sharing the same latent space, autoregressive models significantly lag behind LDMs and MIMs in image generation. This finding contrasts sharply with the field of NLP, where the autoregressive model GPT has established a commanding presence. To address this discrepancy, we introduce a unified perspective on the relationship between latent space and generative models, emphasizing the stability of latent space in image generative modeling. Furthermore, we propose a simple but effective discrete image tokenizer to stabilize the latent space for image generative modeling by applying K-Means on the latent features of self-supervised learning models. Experimental results show that image autoregressive modeling with our tokenizer (DiGIT) benefits both image understanding and image generation with the next token prediction principle, which is inherently straightforward for GPT models but challenging for other generative models. Remarkably, for the first time, a GPT-style autoregressive model for images outperforms LDMs, which also exhibits substantial improvement akin to GPT when scaling up model size. Our findings underscore the potential of an optimized latent space and the integration of discrete tokenization in advancing the capabilities of image generative models. The code is available at \url{https://github.com/DAMO-NLP-SG/DiGIT}.

Songtao Jiang, Sibo Song, Chenyi Zhou et al.

The transition from image to video understanding requires vision-language models (VLMs) to shift from recognizing static patterns to reasoning over temporal dynamics such as motion trajectories, speed changes, and state transitions. Yet current post-training methods fall short due to two critical limitations: (1) existing datasets often lack temporal-centricity, where answers can be inferred from isolated keyframes rather than requiring holistic temporal integration; and (2) training data generated by proprietary models contains systematic errors in fundamental temporal perception, such as confusing motion directions or misjudging speeds. We introduce SynRL, a post-training framework that teaches models temporal primitives, the fundamental building blocks of temporal understanding including direction, speed, and state tracking. Our key insight is that these abstract primitives, learned from programmatically generated synthetic videos, transfer effectively to real-world scenarios. We decompose temporal understanding into short-term perceptual primitives (speed, direction) and long-term cognitive primitives, constructing 7.7K CoT and 7K RL samples with ground-truth frame-level annotations through code-based video generation. Despite training on simple geometric shapes, SynRL achieves substantial improvements across 15 benchmarks spanning temporal grounding, complex reasoning, and general video understanding. Remarkably, our 7.7K synthetic CoT samples outperform Video-R1 with 165K real-world samples. We attribute this to fundamental temporal skills, such as tracking frame by frame changes and comparing velocity, that transfer effectively from abstract synthetic patterns to complex real-world scenarios. This establishes a new paradigm for video post-training: video temporal learning through carefully designed synthetic data provides a more cost efficient scaling path.

Shuai Bai, Keqin Chen, Xuejing Liu et al. · pku

We introduce Qwen2.5-VL, the latest flagship model of Qwen vision-language series, which demonstrates significant advancements in both foundational capabilities and innovative functionalities. Qwen2.5-VL achieves a major leap forward in understanding and interacting with the world through enhanced visual recognition, precise object localization, robust document parsing, and long-video comprehension. A standout feature of Qwen2.5-VL is its ability to localize objects using bounding boxes or points accurately. It provides robust structured data extraction from invoices, forms, and tables, as well as detailed analysis of charts, diagrams, and layouts. To handle complex inputs, Qwen2.5-VL introduces dynamic resolution processing and absolute time encoding, enabling it to process images of varying sizes and videos of extended durations (up to hours) with second-level event localization. This allows the model to natively perceive spatial scales and temporal dynamics without relying on traditional normalization techniques. By training a native dynamic-resolution Vision Transformer (ViT) from scratch and incorporating Window Attention, we reduce computational overhead while maintaining native resolution. As a result, Qwen2.5-VL excels not only in static image and document understanding but also as an interactive visual agent capable of reasoning, tool usage, and task execution in real-world scenarios such as operating computers and mobile devices. Qwen2.5-VL is available in three sizes, addressing diverse use cases from edge AI to high-performance computing. The flagship Qwen2.5-VL-72B model matches state-of-the-art models like GPT-4o and Claude 3.5 Sonnet, particularly excelling in document and diagram understanding. Additionally, Qwen2.5-VL maintains robust linguistic performance, preserving the core language competencies of the Qwen2.5 LLM.

Weitong Cai, Hang Zhang, Yukai Huang et al.

Always-on sensing is essential for next-generation edge/wearable AI systems, yet continuous high-fidelity RGB video capture remains prohibitively expensive for resource-constrained mobile and edge platforms. We present a new paradigm for efficient streaming video understanding: grayscale-always, color-on-demand. Through preliminary studies, we discover that color is not always necessary. Sparse RGB frames suffice for comparable performance when temporal structure is preserved via continuous grayscale streams. Building on this insight, we propose ColorTrigger, an online training-free trigger that selectively activates color capture based on windowed grayscale affinity analysis. Designed for real-time edge deployment, ColorTrigger uses lightweight quadratic programming to detect chromatic redundancy causally, coupled with credit-budgeted control and dynamic token routing to jointly reduce sensing and inference costs. On streaming video understanding benchmarks, ColorTrigger achieves 91.6% of full-color baseline performance while using only 8.1% RGB frames, demonstrating substantial color redundancy in natural videos and enabling practical always-on video sensing on resource-constrained devices.

Chuyue Wang, Jie Feng, Yuxi Wu et al.

Human mobility prediction is a critical task but remains challenging due to its complexity and variability across populations and regions. Recently, large language models (LLMs) have made progress in zero-shot prediction, but existing methods suffer from limited interpretability (due to black-box reasoning), lack of iterative learning from new data, and poor transferability. In this paper, we introduce \textbf{ARMove}, a fully transferable framework for predicting human mobility through agentic reasoning. To address these limitations, ARMove employs standardized feature management with iterative optimization and user-specific customization: four major feature pools for foundational knowledge, user profiles for segmentation, and an automated generation mechanism integrating LLM knowledge. Robust generalization is achieved via agentic decision-making that adjusts feature weights to maximize accuracy while providing interpretable decision paths. Finally, large-small model synergy distills strategies from large LLMs (e.g., 72B) to smaller ones (e.g., 7B), reducing costs and enhancing performance ceilings. Extensive experiments on four global datasets show ARMove outperforms state-of-the-art baselines on 6 out of 12 metrics (gains of 0.78\% to 10.47\%), with transferability tests confirming robustness across regions, users, and scales. The other 4 items also achieved suboptimal results. Transferability tests confirm its 19 robustness across regions, user groups, and model scales, while interpretability 20 analysis highlights its transparency in decision-making. Our codes are available at: https://anonymous.4open.science/r/ARMove-F847.

Hang Zhang

In this research, I proposed a network structure for multi-view 3D object detection using camera-only data and a Bird's-Eye-View map. My work is based on a current key challenge domain adaptation and visual data transfer. Although many excellent camera-only 3D object detection has been continuously proposed, many research work risk dramatic performance drop when the networks are trained on the source domain but tested on a different target domain. Then I found it is very surprising that predictions on bounding boxes and classes are still replied to on 2D networks. Based on the domain gap assumption on various 3D datasets, I found they still shared a similar data extraction on the same BEV map size and camera data transfer. Therefore, to analyze the domain gap influence on the current method and to make good use of 3D space information among the dataset and the real world, I proposed a transfer learning method and Transformer construction to study the 3D object detection on NuScenes-mini and Lyft. Through multi-dataset training and a detection head from the Transformer, the network demonstrated good data migration performance and efficient detection performance by using 3D anchor query and 3D positional information. Relying on only a small amount of source data and the existing large model pre-training weights, the efficient network manages to achieve competitive results on the new target domain. Moreover, my study utilizes 3D information as available semantic information and 2D multi-view image features blending into the visual-language transfer design. In the final 3D anchor box prediction and object classification, my network achieved good results on standard metrics of 3D object detection, which differs from dataset-specific models on each training domain without any fine-tuning.

Wenqi Zhang, Hang Zhang, Xin Li et al.

Compared to image-text pair data, interleaved corpora enable Vision-Language Models (VLMs) to understand the world more naturally like humans. However, such existing datasets are crawled from webpage, facing challenges like low knowledge density, loose image-text relations, and poor logical coherence between images. On the other hand, the internet hosts vast instructional videos (e.g., online geometry courses) that are widely used by humans to learn foundational subjects, yet these valuable resources remain underexplored in VLM training. In this paper, we introduce a high-quality \textbf{multimodal textbook} corpus with richer foundational knowledge for VLM pretraining. It collects over 2.5 years of instructional videos, totaling 22,000 class hours. We first use an LLM-proposed taxonomy to systematically gather instructional videos. Then we progressively extract and refine visual (keyframes), audio (ASR), and textual knowledge (OCR) from the videos, and organize as an image-text interleaved corpus based on temporal order. Compared to its counterparts, our video-centric textbook offers more coherent context, richer knowledge, and better image-text alignment. Experiments demonstrate its superb pretraining performance, particularly in knowledge- and reasoning-intensive tasks like ScienceQA and MathVista. Moreover, VLMs pre-trained on our textbook exhibit outstanding interleaved context awareness, leveraging visual and textual cues in their few-shot context for task solving. Our code are available at https://github.com/DAMO-NLP-SG/multimodal_textbook.

Zheng Zhang, Haonan Li, Xingyu Li et al.

Bug bisection has been an important security task that aims to understand the range of software versions impacted by a bug, i.e., identifying the commit that introduced the bug. However, traditional patch-based bisection methods are faced with several significant barriers: For example, they assume that the bug-inducing commit (BIC) and the patch commit modify the same functions, which is not always true. They often rely solely on code changes, while the commit message frequently contains a wealth of vulnerability-related information. They are also based on simple heuristics (e.g., assuming the BIC initializes lines deleted in the patch) and lack any logical analysis of the vulnerability. In this paper, we make the observation that Large Language Models (LLMs) are well-positioned to break the barriers of existing solutions, e.g., comprehend both textual data and code in patches and commits. Unlike previous BIC identification approaches, which yield poor results, we propose a comprehensive multi-stage pipeline that leverages LLMs to: (1) fully utilize patch information, (2) compare multiple candidate commits in context, and (3) progressively narrow down the candidates through a series of down-selection steps. In our evaluation, we demonstrate that our approach achieves significantly better accuracy than the state-of-the-art solution by more than 38\%. Our results further confirm that the comprehensive multi-stage pipeline is essential, as it improves accuracy by 60\% over a baseline LLM-based bisection method.

Hang Zhang, Xiang Chen, Renjiu Hu et al.

Learning-based deformable image registration (DIR) accelerates alignment by amortizing traditional optimization via neural networks. Label supervision further enhances accuracy, enabling efficient and precise nonlinear alignment of unseen scans. However, images with sparse features amid large smooth regions, such as retinal vessels, introduce aperture and large-displacement challenges that unsupervised DIR methods struggle to address. This limitation occurs because neural networks predict deformation fields in a single forward pass, leaving fields unconstrained post-training and shifting the regularization burden entirely to network weights. To address these issues, we introduce SmoothProper, a plug-and-play neural module enforcing smoothness and promoting message passing within the network's forward pass. By integrating a duality-based optimization layer with tailored interaction terms, SmoothProper efficiently propagates flow signals across spatial locations, enforces smoothness, and preserves structural consistency. It is model-agnostic, seamlessly integrates into existing registration frameworks with minimal parameter overhead, and eliminates regularizer hyperparameter tuning. Preliminary results on a retinal vessel dataset exhibiting aperture and large-displacement challenges demonstrate our method reduces registration error to 1.88 pixels on 2912x2912 images, marking the first unsupervised DIR approach to effectively address both challenges. The source code will be available at https://github.com/tinymilky/SmoothProper.

Siqi Chen, Xinyu Dong, Haolei Xu et al.

Large Language Models (LLMs) and Multimodal LLMs have shown promising capabilities for SVG processing, yet existing benchmarks suffer from limited real-world coverage, lack of complexity stratification, and fragmented evaluation paradigms. We introduce SVGenius, a comprehensive benchmark comprising 2,377 queries across three progressive dimensions: understanding, editing, and generation. Built on real-world data from 24 application domains with systematic complexity stratification, SVGenius evaluates models through 8 task categories and 18 metrics. We assess 22 mainstream models spanning different scales, architectures, training paradigms, and accessibility levels. Our analysis reveals that while proprietary models significantly outperform open-source counterparts, all models exhibit systematic performance degradation with increasing complexity, indicating fundamental limitations in current approaches; however, reasoning-enhanced training proves more effective than pure scaling for overcoming these limitations, though style transfer remains the most challenging capability across all model types. SVGenius establishes the first systematic evaluation framework for SVG processing, providing crucial insights for developing more capable vector graphics models and advancing automated graphic design applications. Appendix and supplementary materials (including all data and code) are available at https://zju-real.github.io/SVGenius.

Hang Zhang, Qian Lou, Yanshan Wang

Large language models (LLMs) are increasingly utilized in healthcare applications. However, their deployment in clinical practice raises significant safety concerns, including the potential spread of harmful information. This study systematically assesses the vulnerabilities of seven LLMs to three advanced black-box jailbreaking techniques within medical contexts. To quantify the effectiveness of these techniques, we propose an automated and domain-adapted agentic evaluation pipeline. Experiment results indicate that leading commercial and open-source LLMs are highly vulnerable to medical jailbreaking attacks. To bolster model safety and reliability, we further investigate the effectiveness of Continual Fine-Tuning (CFT) in defending against medical adversarial attacks. Our findings underscore the necessity for evolving attack methods evaluation, domain-specific safety alignment, and LLM safety-utility balancing. This research offers actionable insights for advancing the safety and reliability of AI clinicians, contributing to ethical and effective AI deployment in healthcare.

Sicong Leng, Jing Wang, Jiaxi Li et al.

Large multimodal reasoning models have achieved rapid progress, but their advancement is constrained by two major limitations: the absence of open, large-scale, high-quality long chain-of-thought (CoT) data, and the instability of reinforcement learning (RL) algorithms in post-training. Group Relative Policy Optimization (GRPO), the standard framework for RL fine-tuning, is prone to gradient vanishing when reward variance is low, which weakens optimization signals and impairs convergence. This work makes three contributions: (1) We propose Variance-Aware Sampling (VAS), a data selection strategy guided by Variance Promotion Score (VPS) that combines outcome variance and trajectory diversity to promote reward variance and stabilize policy optimization. (2) We release large-scale, carefully curated resources containing ~1.6M long CoT cold-start data and ~15k RL QA pairs, designed to ensure quality, difficulty, and diversity, along with a fully reproducible end-to-end training codebase. (3) We open-source a family of multimodal reasoning models in multiple scales, establishing standardized baselines for the community. Experiments across mathematical reasoning benchmarks demonstrate the effectiveness of both the curated data and the proposed VAS. Comprehensive ablation studies and analyses provide further insight into the contributions of each component. In addition, we theoretically establish that reward variance lower-bounds the expected policy gradient magnitude, with VAS serving as a practical mechanism to realize this guarantee. Our code, data, and checkpoints are available at https://github.com/LengSicong/MMR1.

Hang Zhang, Yuxi Zhang, Jiazheng Wang et al.

Recent developments in neural networks have improved deformable image registration (DIR) by amortizing iterative optimization, enabling fast and accurate DIR results. However, learning-based methods often face challenges with limited training data, large deformations, and tend to underperform compared to iterative approaches when label supervision is unavailable. While iterative methods can achieve higher accuracy in such scenarios, they are considerably slower than learning-based methods. To address these limitations, we propose VoxelOpt, a discrete optimization-based DIR framework that combines the strengths of learning-based and iterative methods to achieve a better balance between registration accuracy and runtime. VoxelOpt uses displacement entropy from local cost volumes to measure displacement signal strength at each voxel, which differs from earlier approaches in three key aspects. First, it introduces voxel-wise adaptive message passing, where voxels with lower entropy receives less influence from their neighbors. Second, it employs a multi-level image pyramid with 27-neighbor cost volumes at each level, avoiding exponential complexity growth. Third, it replaces hand-crafted features or contrastive learning with a pretrained foundational segmentation model for feature extraction. In abdominal CT registration, these changes allow VoxelOpt to outperform leading iterative in both efficiency and accuracy, while matching state-of-the-art learning-based methods trained with label supervision. The source code will be available at https://github.com/tinymilky/VoxelOpt

Yuxi Zhang, Xiang Chen, Jiazheng Wang et al.

In this paper, we summarize the methods and experimental results we proposed for Task 2 in the learn2reg 2024 Challenge. This task focuses on unsupervised registration of anatomical structures in brain MRI images between different patients. The difficulty lies in: (1) without segmentation labels, and (2) a large amount of data. To address these challenges, we built an efficient backbone network and explored several schemes to further enhance registration accuracy. Under the guidance of the NCC loss function and smoothness regularization loss function, we obtained a smooth and reasonable deformation field. According to the leaderboard, our method achieved a Dice coefficient of 77.34%, which is 1.4% higher than the TransMorph. Overall, we won second place on the leaderboard for Task 2.

Boqiang Zhang, Kehan Li, Zesen Cheng et al. · pku

In this paper, we propose VideoLLaMA3, a more advanced multimodal foundation model for image and video understanding. The core design philosophy of VideoLLaMA3 is vision-centric. The meaning of "vision-centric" is two-fold: the vision-centric training paradigm and vision-centric framework design. The key insight of our vision-centric training paradigm is that high-quality image-text data is crucial for both image and video understanding. Instead of preparing massive video-text datasets, we focus on constructing large-scale and high-quality image-text datasets. VideoLLaMA3 has four training stages: 1) Vision Encoder Adaptation, which enables vision encoder to accept images of variable resolutions as input; 2) Vision-Language Alignment, which jointly tunes the vision encoder, projector, and LLM with large-scale image-text data covering multiple types (including scene images, documents, charts) as well as text-only data. 3) Multi-task Fine-tuning, which incorporates image-text SFT data for downstream tasks and video-text data to establish a foundation for video understanding. 4) Video-centric Fine-tuning, which further improves the model's capability in video understanding. As for the framework design, to better capture fine-grained details in images, the pretrained vision encoder is adapted to encode images of varying sizes into vision tokens with corresponding numbers, rather than a fixed number of tokens. For video inputs, we reduce the number of vision tokens according to their similarity so that the representation of videos will be more precise and compact. Benefit from vision-centric designs, VideoLLaMA3 achieves compelling performances in both image and video understanding benchmarks.

Yujian Yuan, Changjie Wu, Xinyuan Chang et al.

Large-scale map construction plays a vital role in applications like autonomous driving and navigation systems. Traditional large-scale map construction approaches mainly rely on costly and inefficient special data collection vehicles and labor-intensive annotation processes. While existing satellite-based methods have demonstrated promising potential in enhancing the efficiency and coverage of map construction, they exhibit two major limitations: (1) inherent drawbacks of satellite data (e.g., occlusions, outdatedness) and (2) inefficient vectorization from perception-based methods, resulting in discontinuous and rough roads that require extensive post-processing. This paper presents a novel generative framework, UniMapGen, for large-scale map construction, offering three key innovations: (1) representing lane lines as \textbf{discrete sequence} and establishing an iterative strategy to generate more complete and smooth map vectors than traditional perception-based methods. (2) proposing a flexible architecture that supports \textbf{multi-modal} inputs, enabling dynamic selection among BEV, PV, and text prompt, to overcome the drawbacks of satellite data. (3) developing a \textbf{state update} strategy for global continuity and consistency of the constructed large-scale map. UniMapGen achieves state-of-the-art performance on the OpenSatMap dataset. Furthermore, UniMapGen can infer occluded roads and predict roads missing from dataset annotations. Our code will be released.