Xu Zhao

Xu Zhao, Wenchao Ding, Yongqi An et al.

The recently proposed segment anything model (SAM) has made a significant influence in many computer vision tasks. It is becoming a foundation step for many high-level tasks, like image segmentation, image caption, and image editing. However, its huge computation costs prevent it from wider applications in industry scenarios. The computation mainly comes from the Transformer architecture at high-resolution inputs. In this paper, we propose a speed-up alternative method for this fundamental task with comparable performance. By reformulating the task as segments-generation and prompting, we find that a regular CNN detector with an instance segmentation branch can also accomplish this task well. Specifically, we convert this task to the well-studied instance segmentation task and directly train the existing instance segmentation method using only 1/50 of the SA-1B dataset published by SAM authors. With our method, we achieve a comparable performance with the SAM method at 50 times higher run-time speed. We give sufficient experimental results to demonstrate its effectiveness. The codes and demos will be released at https://github.com/CASIA-IVA-Lab/FastSAM.

Yongqi An, Xu Zhao, Tao Yu et al.

Background subtraction (BGS) aims to extract all moving objects in the video frames to obtain binary foreground segmentation masks. Deep learning has been widely used in this field. Compared with supervised-based BGS methods, unsupervised methods have better generalization. However, previous unsupervised deep learning BGS algorithms perform poorly in sophisticated scenarios such as shadows or night lights, and they cannot detect objects outside the pre-defined categories. In this work, we propose an unsupervised BGS algorithm based on zero-shot object detection called Zero-shot Background Subtraction (ZBS). The proposed method fully utilizes the advantages of zero-shot object detection to build the open-vocabulary instance-level background model. Based on it, the foreground can be effectively extracted by comparing the detection results of new frames with the background model. ZBS performs well for sophisticated scenarios, and it has rich and extensible categories. Furthermore, our method can easily generalize to other tasks, such as abandoned object detection in unseen environments. We experimentally show that ZBS surpasses state-of-the-art unsupervised BGS methods by 4.70% F-Measure on the CDnet 2014 dataset. The code is released at https://github.com/CASIA-IVA-Lab/ZBS.

Yueming Hao, Xu Zhao, Bin Bao et al.

Deep learning (DL) has been a revolutionary technique in various domains. To facilitate the model development and deployment, many deep learning frameworks are proposed, among which PyTorch is one of the most popular solutions. The performance of ecosystem around PyTorch is critically important, which saves the costs of training models and reduces the response time of model inferences. In this paper, we propose TorchBench, a novel benchmark suite to study the performance of PyTorch software stack. Unlike existing benchmark suites, TorchBench encloses many representative models, covering a large PyTorch API surface. TorchBench is able to comprehensively characterize the performance of the PyTorch software stack, guiding the performance optimization across models, PyTorch framework, and GPU libraries. We show two practical use cases of TorchBench. (1) We profile TorchBench to identify GPU performance inefficiencies in PyTorch. We are able to optimize many performance bugs and upstream patches to the official PyTorch repository. (2) We integrate TorchBench into PyTorch continuous integration system. We are able to identify performance regression in multiple daily code checkins to prevent PyTorch repository from introducing performance bugs. TorchBench is open source and keeps evolving.

Kaixin Li, Qisheng Hu, Xu Zhao et al.

Code editing encompasses a variety of pragmatic tasks that developers deal with daily. Despite its relevance and practical usefulness, automatic code editing remains an underexplored area in the evolution of deep learning models, partly due to data scarcity. In this work, we explore the use of Large Language Models (LLMs) to edit code based on user instructions. Evaluated on a novel human-written execution-based benchmark dubbed EditEval, we found current models often struggle to fulfill the instructions. In light of this, we contribute InstructCoder, the first instruction-tuning dataset designed to adapt LLMs for general-purpose code editing, containing high-diversity code-editing tasks such as comment insertion, code optimization, and code refactoring. It consists of over 114,000 instruction-input-output triplets and covers multiple distinct code editing scenarios. The collection process starts with filtered commit data sourced from GitHub Python repositories as seeds. Subsequently, the dataset is systematically expanded through an iterative process, where both seed and generated tasks are used to prompt ChatGPT for more data. Our findings reveal that open-source LLMs fine-tuned on InstructCoder can significantly enhance the accuracy of code edits, exhibiting superior code-editing performance matching advanced proprietary LLMs. The datasets and the source code are publicly available at https://github.com/qishenghu/CodeInstruct.

Shuming Liu, Mengmeng Xu, Chen Zhao et al.

Temporal action detection (TAD) with end-to-end training often suffers from the pain of huge demand for computing resources due to long video duration. In this work, we propose an efficient temporal action detector (ETAD) that can train directly from video frames with extremely low GPU memory consumption. Our main idea is to minimize and balance the heavy computation among features and gradients in each training iteration. We propose to sequentially forward the snippet frame through the video encoder, and backward only a small necessary portion of gradients to update the encoder. To further alleviate the computational redundancy in training, we propose to dynamically sample only a small subset of proposals during training. Moreover, various sampling strategies and ratios are studied for both the encoder and detector. ETAD achieves state-of-the-art performance on TAD benchmarks with remarkable efficiency. On ActivityNet-1.3, training ETAD in 18 hours can reach 38.25% average mAP with only 1.3 GB memory consumption per video under end-to-end training. Our code will be publicly released.

Jiaran Zhang, Luck Ma, Fanqi Wan et al.

Reliable Docker-based environment construction is a dominant bottleneck for scaling execution-grounded training and evaluation of software engineering agents. We introduce DockSmith, a specialized agentic Docker builder designed to address this challenge. DockSmith treats environment construction not only as a preprocessing step, but as a core agentic capability that exercises long-horizon tool use, dependency reasoning, and failure recovery, yielding supervision that transfers beyond Docker building itself. DockSmith is trained on large-scale, execution-grounded Docker-building trajectories produced by a SWE-Factory-style pipeline augmented with a loop-detection controller and a cross-task success memory. Training a 30B-A3B model on these trajectories achieves open-source state-of-the-art performance on Multi-Docker-Eval, with 39.72% Fail-to-Pass and 58.28% Commit Rate. Moreover, DockSmith improves out-of-distribution performance on SWE-bench Verified, SWE-bench Multilingual, and Terminal-Bench 2.0, demonstrating broader agentic benefits of environment construction.

Ke Liao, Wei Wang, Armagan Elibol et al.

Machine learning has been widely used in healthcare applications to approximate complex models, for clinical diagnosis, prognosis, and treatment. As deep learning has the outstanding ability to extract information from time series, its true capabilities on sparse, irregularly sampled, multivariate, and imbalanced physiological data are not yet fully explored. In this paper, we systematically examine the performance of machine learning models for the clinical prediction task based on the EHR, especially physiological time series. We choose Physionet 2019 challenge public dataset to predict Sepsis outcomes in ICU units. Ten baseline machine learning models are compared, including 3 deep learning methods and 7 non-deep learning methods, commonly used in the clinical prediction domain. Nine evaluation metrics with specific clinical implications are used to assess the performance of models. Besides, we sub-sample training dataset sizes and use learning curve fit to investigate the impact of the training dataset size on the performance of the machine learning models. We also propose the general pre-processing method for the physiology time-series data and use Dice Loss to deal with the dataset imbalanced problem. The results show that deep learning indeed outperforms non-deep learning, but with certain conditions: firstly, evaluating with some particular evaluation metrics (AUROC, AUPRC, Sensitivity, and FNR), but not others; secondly, the training dataset size is large enough (with an estimation of a magnitude of thousands).

Xu Zhao, Xiting Wang, Weiran Shen

Large language models (LLMs) have achieved impressive capabilities, yet ensuring their safety against harmful prompts remains a critical challenge. Recent work has revealed that the latent representations (embeddings) of harmful and safe queries in LLMs typically exhibit linear separability, a property that has been exploited to construct attacks by perturbing the embeddings of harmful queries towards the safe subspace. Motivated by this observation, we propose a representation-level fine-tuning approach, named Embedding Space Separation (ES2), which improves LLM safety by explicitly enlarging the distance between harmful and safe representations in the embedding space. To prevent degradation of model's general capabilities, we introduce a Kullback-Leibler (KL) divergence regularization term into the loss function, which constrains the logits of the fine-tuned model to align with those of the original base model on harmless inputs. We evaluate our method on several open-source LLMs using standard safety benchmarks. Extensive experimental results demonstrate that our approach substantially improves model safety while maintaining comparable general capabilities.

Zhaowen Li, Xu Zhao, Chaoyang Zhao et al.

Previous unsupervised domain adaptation methods did not handle the cross-domain problem from the perspective of frequency for computer vision. The images or feature maps of different domains can be decomposed into the low-frequency component and high-frequency component. This paper proposes the assumption that low-frequency information is more domain-invariant while the high-frequency information contains domain-related information. Hence, we introduce an approach, named low-frequency module (LFM), to extract domain-invariant feature representations. The LFM is constructed with the digital Gaussian low-pass filter. Our method is easy to implement and introduces no extra hyperparameter. We design two effective ways to utilize the LFM for domain adaptation, and our method is complementary to other existing methods and formulated as a plug-and-play unit that can be combined with these methods. Experimental results demonstrate that our LFM outperforms state-of-the-art methods for various computer vision tasks, including image classification and object detection.

Xiaojing Chen, Jingqi Cheng, Xu Zhao et al.

Deep learning-based Major Depressive Disorder (MDD) detection using Electroencephalography (EEG) is fundamentally constrained by the "small-sample dilemma." Prevailing generative data augmentation methods not only incur heavy computational overhead but also risk introducing synthetic noise, thereby blurring classification boundaries. To challenge the traditional "data quantity first" convention, we propose a novel framework "Beyond Augmentation": Score-Guided Classification (SGC). SGC does not synthesize pseudo-samples; instead, it utilizes an unsupervised generative network architecture to model the structural and statistical anomaly degrees of samples, serving as the core "Pathological Prior". This prior, after robust normalization, is explicitly fused with deep feature representations, thereby precisely guiding the classifier's decision boundary. Furthermore, to dynamically adapt to varying channel configurations, we propose a Cross-Channel Spatial Adaptation module, utilizing a spatial mapping mechanism to effectively resolve the hardware heterogeneity of mismatched channels in multi-center datasets. Extensive experiments on the Mumtaz2016 and high-density MODMA datasets demonstrate the effectiveness and exceptional generalizability of our method under the challenging "zero data augmentation" setting and at "zero sample synthesis cost". Keywords: Electroencephalography (EEG), Depression Detection, Anomaly Score, Diffusion Models, Few-Shot Learning

Zhaowen Li, Xu Zhao, Peigeng Ding et al.

Recent researches indicate that utilizing the frequency information of input data can enhance the performance of networks. However, the existing popular convolutional structure is not designed specifically for utilizing the frequency information contained in datasets. In this paper, we propose a novel and effective module, named FreConv (frequency branch-and-integration convolution), to replace the vanilla convolution. FreConv adopts a dual-branch architecture to extract and integrate high- and low-frequency information. In the high-frequency branch, a derivative-filter-like architecture is designed to extract the high-frequency information while a light extractor is employed in the low-frequency branch because the low-frequency information is usually redundant. FreConv is able to exploit the frequency information of input data in a more reasonable way to enhance feature representation ability and reduce the memory and computational cost significantly. Without any bells and whistles, experimental results on various tasks demonstrate that FreConv-equipped networks consistently outperform state-of-the-art baselines.

Yiming Bao, Xu Zhao, Dahong Qian

There exist challenging problems in 3D human pose estimation mission, such as poor performance caused by occlusion and self-occlusion. Recently, IMU-vision sensor fusion is regarded as valuable for solving these problems. However, previous researches on the fusion of IMU and vision data, which is heterogeneous, fail to adequately utilize either IMU raw data or reliable high-level vision features. To facilitate a more efficient sensor fusion, in this work we propose a framework called \emph{FusePose} under a parametric human kinematic model. Specifically, we aggregate different information of IMU or vision data and introduce three distinctive sensor fusion approaches: NaiveFuse, KineFuse and AdaDeepFuse. NaiveFuse servers as a basic approach that only fuses simplified IMU data and estimated 3D pose in euclidean space. While in kinematic space, KineFuse is able to integrate the calibrated and aligned IMU raw data with converted 3D pose parameters. AdaDeepFuse further develops this kinematical fusion process to an adaptive and end-to-end trainable manner. Comprehensive experiments with ablation studies demonstrate the rationality and superiority of the proposed framework. The performance of 3D human pose estimation is improved compared to the baseline result. On Total Capture dataset, KineFuse surpasses previous state-of-the-art which uses IMU only for testing by 8.6\%. AdaDeepFuse surpasses state-of-the-art which uses IMU for both training and testing by 8.5\%. Moreover, we validate the generalization capability of our framework through experiments on Human3.6M dataset.

Ailin Huang, Ang Li, Aobo Kong et al.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

Yongqi An, Xu Zhao, Tao Yu et al.

Network Pruning is a promising way to address the huge computing resource demands of the deployment and inference of Large Language Models (LLMs). Retraining-free is important for LLMs' pruning methods. However, almost all of the existing retraining-free pruning approaches for LLMs focus on unstructured pruning, which requires specific hardware support for acceleration. In this paper, we propose a novel retraining-free structured pruning framework for LLMs, named FLAP (FLuctuation-based Adaptive Structured Pruning). It is hardware-friendly by effectively reducing storage and enhancing inference speed. For effective structured pruning of LLMs, we highlight three critical elements that demand the utmost attention: formulating structured importance metrics, adaptively searching the global compressed model, and implementing compensation mechanisms to mitigate performance loss. First, FLAP determines whether the output feature map is easily recoverable when a column of weight is removed, based on the fluctuation pruning metric. Then it standardizes the importance scores to adaptively determine the global compressed model structure. At last, FLAP adds additional bias terms to recover the output feature maps using the baseline values. We thoroughly evaluate our approach on a variety of language benchmarks. Without any retraining, our method significantly outperforms the state-of-the-art methods, including LLM-Pruner and the extension of Wanda in structured pruning. The code is released at https://github.com/CASIA-IVA-Lab/FLAP.

Guoqing Ma, Haoyang Huang, Kun Yan et al.

We present Step-Video-T2V, a state-of-the-art text-to-video pre-trained model with 30B parameters and the ability to generate videos up to 204 frames in length. A deep compression Variational Autoencoder, Video-VAE, is designed for video generation tasks, achieving 16x16 spatial and 8x temporal compression ratios, while maintaining exceptional video reconstruction quality. User prompts are encoded using two bilingual text encoders to handle both English and Chinese. A DiT with 3D full attention is trained using Flow Matching and is employed to denoise input noise into latent frames. A video-based DPO approach, Video-DPO, is applied to reduce artifacts and improve the visual quality of the generated videos. We also detail our training strategies and share key observations and insights. Step-Video-T2V's performance is evaluated on a novel video generation benchmark, Step-Video-T2V-Eval, demonstrating its state-of-the-art text-to-video quality when compared with both open-source and commercial engines. Additionally, we discuss the limitations of current diffusion-based model paradigm and outline future directions for video foundation models. We make both Step-Video-T2V and Step-Video-T2V-Eval available at https://github.com/stepfun-ai/Step-Video-T2V. The online version can be accessed from https://yuewen.cn/videos as well. Our goal is to accelerate the innovation of video foundation models and empower video content creators.

Ailin Huang, Boyong Wu, Bruce Wang et al.

Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.

Xiaoyue Wan, Zhuo Chen, Xu Zhao

The rapid development of multi-view 3D human pose estimation (HPE) is attributed to the maturation of monocular 2D HPE and the geometry of 3D reconstruction. However, 2D detection outliers in occluded views due to neglect of view consistency, and 3D implausible poses due to lack of pose coherence, remain challenges. To solve this, we introduce a Multi-View Fusion module to refine 2D results by establishing view correlations. Then, Holistic Triangulation is proposed to infer the whole pose as an entirety, and anatomy prior is injected to maintain the pose coherence and improve the plausibility. Anatomy prior is extracted by PCA whose input is skeletal structure features, which can factor out global context and joint-by-joint relationship from abstract to concrete. Benefiting from the closed-form solution, the whole framework is trained end-to-end. Our method outperforms the state of the art in both precision and plausibility which is assessed by a new metric.

Xiantao Hu, Ying Tai, Xu Zhao et al.

Multimodal tracking has garnered widespread attention as a result of its ability to effectively address the inherent limitations of traditional RGB tracking. However, existing multimodal trackers mainly focus on the fusion and enhancement of spatial features or merely leverage the sparse temporal relationships between video frames. These approaches do not fully exploit the temporal correlations in multimodal videos, making it difficult to capture the dynamic changes and motion information of targets in complex scenarios. To alleviate this problem, we propose a unified multimodal spatial-temporal tracking approach named STTrack. In contrast to previous paradigms that solely relied on updating reference information, we introduced a temporal state generator (TSG) that continuously generates a sequence of tokens containing multimodal temporal information. These temporal information tokens are used to guide the localization of the target in the next time state, establish long-range contextual relationships between video frames, and capture the temporal trajectory of the target. Furthermore, at the spatial level, we introduced the mamba fusion and background suppression interactive (BSI) modules. These modules establish a dual-stage mechanism for coordinating information interaction and fusion between modalities. Extensive comparisons on five benchmark datasets illustrate that STTrack achieves state-of-the-art performance across various multimodal tracking scenarios. Code is available at: https://github.com/NJU-PCALab/STTrack.

Yesheng Zhang, Shuhan Shen, Xu Zhao

We propose MESA and DMESA as novel feature matching methods, which utilize Segment Anything Model (SAM) to effectively mitigate matching redundancy. The key insight of our methods is to establish implicit-semantic area matching prior to point matching, based on advanced image understanding of SAM. Then, informative area matches with consistent internal semantic are able to undergo dense feature comparison, facilitating precise inside-area point matching. Specifically, MESA adopts a sparse matching framework and first obtains candidate areas from SAM results through a novel Area Graph (AG). Then, area matching among the candidates is formulated as graph energy minimization and solved by graphical models derived from AG. To address the efficiency issue of MESA, we further propose DMESA as its dense counterpart, applying a dense matching framework. After candidate areas are identified by AG, DMESA establishes area matches through generating dense matching distributions. The distributions are produced from off-the-shelf patch matching utilizing the Gaussian Mixture Model and refined via the Expectation Maximization. With less repetitive computation, DMESA showcases a speed improvement of nearly five times compared to MESA, while maintaining competitive accuracy. Our methods are extensively evaluated on five datasets encompassing indoor and outdoor scenes. The results illustrate consistent performance improvements from our methods for five distinct point matching baselines across all datasets. Furthermore, our methods exhibit promise generalization and improved robustness against image resolution variations. The code is publicly available at https://github.com/Easonyesheng/A2PM-MESA.

Wenke Xia, Xu Zhao, Xincheng Pang et al.

The imbalance problem is widespread in the field of machine learning, which also exists in multimodal learning areas caused by the intrinsic discrepancy between modalities of samples. Recent works have attempted to solve the modality imbalance problem from algorithm perspective, however, they do not fully analyze the influence of modality bias in datasets. Concretely, existing multimodal datasets are usually collected under specific tasks, where one modality tends to perform better than other ones in most conditions. In this work, to comprehensively explore the influence of modality bias, we first split existing datasets into different subsets by estimating sample-wise modality discrepancy. We surprisingly find that: the multimodal models with existing imbalance algorithms consistently perform worse than the unimodal one on specific subsets, in accordance with the modality bias. To further explore the influence of modality bias and analyze the effectiveness of existing imbalance algorithms, we build a balanced audiovisual dataset, with uniformly distributed modality discrepancy over the whole dataset. We then conduct extensive experiments to re-evaluate existing imbalance algorithms and draw some interesting findings: existing algorithms only provide a compromise between modalities and suffer from the large modality discrepancy of samples. We hope that these findings could facilitate future research on the modality imbalance problem.

Boyong Wu, Chao Yan, Chen Hu et al.

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

Yesheng Zhang, Xu Zhao

Feature matching is a crucial technique in computer vision. A unified perspective for this task is to treat it as a searching problem, aiming at an efficient search strategy to narrow the search space to point matches between images. One of the key aspects of search strategy is the search space, which in current approaches is not carefully defined, resulting in limited matching accuracy. This paper, thus, pays attention to the search space and proposes to set the initial search space for point matching as the matched image areas containing prominent semantic, named semantic area matches. This search space favors point matching by salient features and alleviates the accuracy limitation in recent Transformer-based matching methods. To achieve this search space, we introduce a hierarchical feature matching framework: Area to Point Matching (A2PM), to first find semantic area matches between images and later perform point matching on area matches. We further propose Semantic and Geometry Area Matching (SGAM) method to realize this framework, which utilizes semantic prior and geometry consistency to establish accurate area matches between images. By integrating SGAM with off-the-shelf state-of-the-art matchers, our method, adopting the A2PM framework, achieves encouraging precision improvements in massive point matching and pose estimation experiments.

Xiaodong Wang, Langling Huang, Zhirong Wu et al.

The development of multimodal large language models (MLLMs) has advanced general video understanding. However, existing video evaluation benchmarks primarily focus on non-interactive videos, such as movies and recordings. To fill this gap, this paper proposes the first omnimodal benchmark for interactive livestream videos, LiViBench. It features a diverse set of 24 tasks, highlighting the perceptual, reasoning, and livestream-specific challenges. To efficiently construct the dataset, we design a standardized semi-automatic annotation workflow that incorporates the human-in-the-loop at multiple stages. The workflow leverages multiple MLLMs to form a multi-agent system for comprehensive video description and uses a seed-question-driven method to construct high-quality annotations. All interactive videos in the benchmark include audio, speech, and real-time comments modalities. To enhance models' understanding of interactive videos, we design tailored two-stage instruction-tuning and propose a Video-to-Comment Retrieval (VCR) module to improve the model's ability to utilize real-time comments. Based on these advancements, we develop LiVi-LLM-7B, an MLLM with enhanced knowledge of interactive livestreams. Experiments show that our model outperforms larger open-source models with up to 72B parameters, narrows the gap with leading proprietary models on LiViBench, and achieves enhanced performance on general video benchmarks, including VideoMME, LongVideoBench, MLVU, and VideoEval-Pro.

Xin He, Haiyun Guo, Kuan Zhu et al.

Lane detection plays an important role in autonomous driving perception systems. As deep learning algorithms gain popularity, monocular lane detection methods based on them have demonstrated superior performance and emerged as a key research direction in autonomous driving perception. The core designs of these algorithmic frameworks can be summarized as follows: (1) Task paradigm, focusing on lane instance-level discrimination; (2) Lane modeling, representing lanes as a set of learnable parameters in the neural network; (3) Global context supplementation, enhancing inference on the obscure lanes; (4) Perspective effect elimination, providing accurate 3D lanes for downstream applications. From these perspectives, this paper presents a comprehensive overview of existing methods, encompassing both the increasingly mature 2D lane detection approaches and the developing 3D lane detection works. Besides, this paper compares the performance of mainstream methods on different benchmarks and investigates their inference speed under a unified setting for fair comparison. Moreover, we present some extended works on lane detection, including multi-task perception, video lane detection, online high-definition map construction, and lane topology reasoning, to offer readers a comprehensive roadmap for the evolution of lane detection. Finally, we point out some potential future research directions in this field. We exhaustively collect the papers and codes of existing works at https://github.com/Core9724/Awesome-Lane-Detection and will keep tracing the research.

Yongqi An, Xu Zhao, Tao Yu et al.

Outliers have been widely observed in Large Language Models (LLMs), significantly impacting model performance and posing challenges for model compression. Understanding the functionality and formation mechanisms of these outliers is critically important. Existing works, however, largely focus on reducing the impact of outliers from an algorithmic perspective, lacking an in-depth investigation into their causes and roles. In this work, we provide a detailed analysis of the formation process, underlying causes, and functions of outliers in LLMs. We define and categorize three types of outliers-activation outliers, weight outliers, and attention outliers-and analyze their distributions across different dimensions, uncovering inherent connections between their occurrences and their ultimate influence on the attention mechanism. Based on these observations, we hypothesize and explore the mechanisms by which these outliers arise and function, demonstrating through theoretical derivations and experiments that they emerge due to the self-attention mechanism's softmax operation. These outliers act as implicit context-aware scaling factors within the attention mechanism. As these outliers stem from systematic influences, we term them systematic outliers. Our study not only enhances the understanding of Transformer-based LLMs but also shows that structurally eliminating outliers can accelerate convergence and improve model compression. The code is avilable at https://github.com/an-yongqi/systematic-outliers.

Bin Xia, Bohao Peng, Jiyang Liu et al.

Recently unified generation and editing models have achieved remarkable success with their impressive performance. These models rely mainly on text prompts for instruction-based editing and generation, but language often fails to capture users intended edit locations and fine-grained visual details. To this end, we propose two tasks: scribble-based editing and generation, that enables more flexible creation on graphical user interface (GUI) combining user textual, images, and freehand sketches. We introduce DreamOmni3, tackling two challenges: data creation and framework design. Our data synthesis pipeline includes two parts: scribble-based editing and generation. For scribble-based editing, we define four tasks: scribble and instruction-based editing, scribble and multimodal instruction-based editing, image fusion, and doodle editing. Based on DreamOmni2 dataset, we extract editable regions and overlay hand-drawn boxes, circles, doodles or cropped image to construct training data. For scribble-based generation, we define three tasks: scribble and instruction-based generation, scribble and multimodal instruction-based generation, and doodle generation, following similar data creation pipelines. For the framework, instead of using binary masks, which struggle with complex edits involving multiple scribbles, images, and instructions, we propose a joint input scheme that feeds both the original and scribbled source images into the model, using different colors to distinguish regions and simplify processing. By applying the same index and position encodings to both images, the model can precisely localize scribbled regions while maintaining accurate editing. Finally, we establish comprehensive benchmarks for these tasks to promote further research. Experimental results demonstrate that DreamOmni3 achieves outstanding performance, and models and code will be publicly released.

Xiaoyue Wan, Zhuo Chen, Yiming Bao et al.

In the domain of 3D Human Pose Estimation, which finds widespread daily applications, the requirement for convenient acquisition equipment continues to grow. To satisfy this demand, we set our sights on a short-baseline binocular setting that offers both portability and a geometric measurement property that radically mitigates depth ambiguity. However, as the binocular baseline shortens, two serious challenges emerge: first, the robustness of 3D reconstruction against 2D errors deteriorates; and second, occlusion reoccurs due to the limited visual differences between two views. To address the first challenge, we propose the Stereo Co-Keypoints Estimation module to improve the view consistency of 2D keypoints and enhance the 3D robustness. In this module, the disparity is utilized to represent the correspondence of binocular 2D points and the Stereo Volume Feature is introduced to contain binocular features across different disparities. Through the regression of SVF, two-view 2D keypoints are simultaneously estimated in a collaborative way which restricts their view consistency. Furthermore, to deal with occlusions, a Pre-trained Pose Transformer module is introduced. Through this module, 3D poses are refined by perceiving pose coherence, a representation of joint correlations. This perception is injected by the Pose Transformer network and learned through a pre-training task that recovers iterative masked joints. Comprehensive experiments carried out on H36M and MHAD datasets, complemented by visualizations, validate the effectiveness of our approach in the short-baseline binocular 3D Human Pose Estimation and occlusion handling.

Zhiqian Zhang, Xu Zhao, Xiaoqing Xu et al.

In recent years, multimodal large models have continued to improve on general benchmarks. However, in real-world content moderation and adversarial settings, mainstream models still suffer from degraded generalization and catastrophic forgetting because of limited fine-grained visual perception and insufficient modeling of long-tail noise. In this paper, we present Xuanwu VL-2B as a case study of how general multimodal models can be developed into an industrial-grade foundation model for content ecosystems. The model adopts a compact InternViT-300M + MLP + Qwen3 1.7B architecture, balancing fine-grained visual perception, language-semantic alignment, and deployment cost within an approximately 2B-parameter budget. To balance business specialization with the retention of general capabilities, we developed a data iteration and curation mechanism and trained the model through a progressive three-stage pipeline: pre-training, mid-training, and post-training. Ablation studies and offline business evaluations show that Xuanwu VL-2B achieves an average score of 67.90 across seven OpenCompass multimodal metrics (vs. 64.27 for InternVL 3.5 2B), an average recall of 94.38% over seven independent business moderation tasks, and a weighted overall recall of 82.82% on policy-violating text in challenging adversarial OCR scenarios, outperforming Gemini-2.5-Pro (76.72%). These results show that, under a limited parameter budget, Xuanwu VL-2B achieves a practical balance among business alignment, visual perception, general capability retention, and deployment cost.

Guo Ye, Zexi Zhang, Xu Zhao et al.

Vision-Language-Action (VLA) models have shown remarkable generalization by mapping web-scale knowledge to robotic control, yet they remain blind to physical contact. Consequently, they struggle with contact-rich manipulation tasks that require reasoning about force, texture, and slip. While some approaches incorporate low-dimensional tactile signals, they fail to capture the high-resolution dynamics essential for such interactions. To address this limitation, we introduce DreamTacVLA, a framework that grounds VLA models in contact physics by learning to feel the future. Our model adopts a hierarchical perception scheme in which high-resolution tactile images serve as micro-vision inputs coupled with wrist-camera local vision and third-person macro vision. To reconcile these multi-scale sensory streams, we first train a unified policy with a Hierarchical Spatial Alignment (HSA) loss that aligns tactile tokens with their spatial counterparts in the wrist and third-person views. To further deepen the model's understanding of fine-grained contact dynamics, we finetune the system with a tactile world model that predicts future tactile signals. To mitigate tactile data scarcity and the wear-prone nature of tactile sensors, we construct a hybrid large-scale dataset sourced from both high-fidelity digital twin and real-world experiments. By anticipating upcoming tactile states, DreamTacVLA acquires a rich model of contact physics and conditions its actions on both real observations and imagined consequences. Across contact-rich manipulation tasks, it outperforms state-of-the-art VLA baselines, achieving up to 95% success, highlighting the importance of understanding physical contact for robust, touch-aware robotic agents.

Ziqi Gao, Yaotian Zhu, Qingcheng Zeng et al.

Probing studies what information is encoded in a frozen LLM's layer representations by training a lightweight predictor on top of them. Beyond analysis, probes are often used operationally in probe-then-steer pipelines: a learned concept vector is extracted from a probe and injected via additive activation steering by adding it to a layer representation during the forward pass. The effectiveness of this pipeline hinges on estimating concept vectors that are accurate, directionally stable under ablation, and inexpensive to obtain. Motivated by these desiderata, we propose RAPTOR (Ridge-Adaptive Logistic Probe), a simple L2-regularized logistic probe whose validation-tuned ridge strength yields concept vectors from normalized weights. Across extensive experiments on instruction-tuned LLMs and human-written concept datasets, RAPTOR matches or exceeds strong baselines in accuracy while achieving competitive directional stability and substantially lower training cost; these quantitative results are supported by qualitative downstream steering demonstrations. Finally, using the Convex Gaussian Min-max Theorem (CGMT), we provide a mechanistic characterization of ridge logistic regression in an idealized Gaussian teacher-student model in the high-dimensional few-shot regime, explaining how penalty strength mediates probe accuracy and concept-vector stability and yielding structural predictions that qualitatively align with trends observed on real LLM embeddings.

Yesheng Zhang, Xu Zhao, Dahong Qian

Camera calibration is a crucial technique which significantly influences the performance of many robotic systems. Robustness and high precision have always been the pursuit of diverse calibration methods. State-of-the-art calibration techniques based on classical Zhang's method, however, still suffer from environmental noise, radial lens distortion and sub-optimal parameter estimation. Therefore, in this paper, we propose a hybrid camera calibration framework which combines learning-based approaches with traditional methods to handle these bottlenecks. In particular, this framework leverages learning-based approaches to perform efficient distortion correction and robust chessboard corner coordinate encoding. For sub-pixel accuracy of corner detection, a specially-designed coordinate decoding algorithm with embed outlier rejection mechanism is proposed. To avoid sub-optimal estimation results, we improve the traditional parameter estimation by RANSAC algorithm and achieve stable results. Compared with two widely-used camera calibration toolboxes, experiment results on both real and synthetic datasets manifest the better robustness and higher precision of the proposed framework. The massive synthetic dataset is the basis of our framework's decent performance and will be publicly available along with the code at https://github.com/Easonyesheng/CCS.

Nanfei Jiang, Xu Zhao, Chaoyang Zhao et al.

Structural neural network pruning aims to remove the redundant channels in the deep convolutional neural networks (CNNs) by pruning the filters of less importance to the final output accuracy. To reduce the degradation of performance after pruning, many methods utilize the loss with sparse regularization to produce structured sparsity. In this paper, we analyze these sparsity-training-based methods and find that the regularization of unpruned channels is unnecessary. Moreover, it restricts the network's capacity, which leads to under-fitting. To solve this problem, we propose a novel pruning method, named MaskSparsity, with pruning-aware sparse regularization. MaskSparsity imposes the fine-grained sparse regularization on the specific filters selected by a pruning mask, rather than all the filters of the model. Before the fine-grained sparse regularization of MaskSparity, we can use many methods to get the pruning mask, such as running the global sparse regularization. MaskSparsity achieves 63.03%-FLOPs reduction on ResNet-110 by removing 60.34% of the parameters, with no top-1 accuracy loss on CIFAR-10. On ILSVRC-2012, MaskSparsity reduces more than 51.07% FLOPs on ResNet-50, with only a loss of 0.76% in the top-1 accuracy. The code is released at https://github.com/CASIA-IVA-Lab/MaskSparsity. Moreover, we have integrated the code of MaskSparity into a PyTorch pruning toolkit, EasyPruner, at https://gitee.com/casia_iva_engineer/easypruner.

Shenzheng Zhang, Qi Tan, Xinzhi Zheng et al.

The gap between the randomly initialized item ID embedding and the well-trained warm item ID embedding makes the cold items hard to suit the recommendation system, which is trained on the data of historical warm items. To alleviate the performance decline of new items recommendation, the distribution of the new item ID embedding should be close to that of the historical warm items. To achieve this goal, we propose an Adversarial Variational Auto-encoder Warm-up model (AVAEW) to generate warm-up item ID embedding for cold items. Specifically, we develop a conditional variational auto-encoder model to leverage the side information of items for generating the warm-up item ID embedding. Particularly, we introduce an adversarial module to enforce the alignment between warm-up item ID embedding distribution and historical item ID embedding distribution. We demonstrate the effectiveness and compatibility of the proposed method by extensive offline experiments on public datasets and online A/B tests on a real-world large-scale news recommendation platform.

Haoliang Liu, Chengkun Cai, Xu Zhao et al.

Electroencephalography (EEG) signals are frequently contaminated by artifacts, affecting the accuracy of subsequent analysis. Traditional artifact removal methods are often computationally expensive and inefficient for real-time applications in edge devices. This paper presents a method to reduce the computational cost of most existing convolutional neural networks (CNN) by replacing one-dimensional (1-D) CNNs with two-dimensional (2-D) CNNs and deploys them on Edge Tensor Processing Unit (TPU), which is an open-resource hardware accelerator widely used in edge devices for low-latency, low-power operation. A new Efficient 2D-CNN Artifact Removal (E2CAR) framework is also represented using the method above, and it achieves a 90\% reduction in inference time on the TPU and decreases power consumption by 18.98\%, while maintaining comparable artifact removal performance to existing methods. This approach facilitates efficient EEG signal processing on edge devices.

Yesheng Zhang, Xu Zhao

Feature matching is a crucial task in the field of computer vision, which involves finding correspondences between images. Previous studies achieve remarkable performance using learning-based feature comparison. However, the pervasive presence of matching redundancy between images gives rise to unnecessary and error-prone computations in these methods, imposing limitations on their accuracy. To address this issue, we propose MESA, a novel approach to establish precise area (or region) matches for efficient matching redundancy reduction. MESA first leverages the advanced image understanding capability of SAM, a state-of-the-art foundation model for image segmentation, to obtain image areas with implicit semantic. Then, a multi-relational graph is proposed to model the spatial structure of these areas and construct their scale hierarchy. Based on graphical models derived from the graph, the area matching is reformulated as an energy minimization task and effectively resolved. Extensive experiments demonstrate that MESA yields substantial precision improvement for multiple point matchers in indoor and outdoor downstream tasks, e.g. +13.61% for DKM in indoor pose estimation.

StepFun, Bin Wang, Bojun Wang et al.

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

Chengkun Cai, Haoliang Liu, Xu Zhao et al.

In the rapidly evolving field of image generation, achieving precise control over generated content and maintaining semantic consistency remain significant limitations, particularly concerning grounding techniques and the necessity for model fine-tuning. To address these challenges, we propose BayesGenie, an off-the-shelf approach that integrates Large Language Models (LLMs) with Bayesian Optimization to facilitate precise and user-friendly image editing. Our method enables users to modify images through natural language descriptions without manual area marking, while preserving the original image's semantic integrity. Unlike existing techniques that require extensive pre-training or fine-tuning, our approach demonstrates remarkable adaptability across various LLMs through its model-agnostic design. BayesGenie employs an adapted Bayesian optimization strategy to automatically refine the inference process parameters, achieving high-precision image editing with minimal user intervention. Through extensive experiments across diverse scenarios, we demonstrate that our framework significantly outperforms existing methods in both editing accuracy and semantic preservation, as validated using different LLMs including Claude3 and GPT-4.

Ailin Huang, Bingxin Li, Bruce Wang et al.

Large Audio-Language Models (LALMs) have significantly advanced intelligent human-computer interaction, yet their reliance on text-based outputs limits their ability to generate natural speech responses directly, hindering seamless audio interactions. To address this, we introduce Step-Audio-AQAA, a fully end-to-end LALM designed for Audio Query-Audio Answer (AQAA) tasks. The model integrates a dual-codebook audio tokenizer for linguistic and semantic feature extraction, a 130-billion-parameter backbone LLM and a neural vocoder for high-fidelity speech synthesis. Our post-training approach employs interleaved token-output of text and audio to enhance semantic coherence and combines Direct Preference Optimization (DPO) with model merge to improve performance. Evaluations on the StepEval-Audio-360 benchmark demonstrate that Step-Audio-AQAA excels especially in speech control, outperforming the state-of-art LALMs in key areas. This work contributes a promising solution for end-to-end LALMs and highlights the critical role of token-based vocoder in enhancing overall performance for AQAA tasks.

Xu Zhao, Chen Zhao, Xiantao Hu et al.

Recent advancements in multi-scale architectures have demonstrated exceptional performance in image denoising tasks. However, existing architectures mainly depends on a fixed single-input single-output Unet architecture, ignoring the multi-scale representations of pixel level. In addition, previous methods treat the frequency domain uniformly, ignoring the different characteristics of high-frequency and low-frequency noise. In this paper, we propose a novel multi-scale adaptive dual-domain network (MADNet) for image denoising. We use image pyramid inputs to restore noise-free results from low-resolution images. In order to realize the interaction of high-frequency and low-frequency information, we design an adaptive spatial-frequency learning unit (ASFU), where a learnable mask is used to separate the information into high-frequency and low-frequency components. In the skip connections, we design a global feature fusion block to enhance the features at different scales. Extensive experiments on both synthetic and real noisy image datasets verify the effectiveness of MADNet compared with current state-of-the-art denoising approaches.

Chengkun Cai, Xu Zhao, Yucheng Du et al.

Large Language Models (LLMs) have emerged as powerful tools in artificial intelligence, especially in complex decision-making scenarios, but their static problem-solving strategies often limit their adaptability to dynamic environments. We explore the enhancement of reasoning capabilities in LLMs through Temperature Tree ($T^2$) prompting via a heuristic algorithm, termed as $T^2$ of Thoughts ($T^2oT$). The primary focus is on enhancing decision-making processes by dynamically adjusting search parameters, especially temperature, to improve accuracy without increasing computational demands. We empirically validate that our hybrid $T^2oT$ approach yields enhancements in, single-solution accuracy, multi-solution generation and text generation quality. Our findings suggest that while dynamic search depth adjustments based on temperature can yield mixed results, a fixed search depth, when coupled with adaptive capabilities of $T^2oT$, provides a more reliable and versatile problem-solving strategy. This work highlights the potential for future explorations in optimizing algorithmic interactions with foundational language models, particularly illustrated by our development for the Game of 24 and Creative Writing tasks.

Baoxing Li, Yong Deng, Yehui Yang et al.

To reconstruct a 3D human surface from a single image, it is crucial to simultaneously consider human pose, shape, and clothing details. Recent approaches have combined parametric body models (such as SMPL), which capture body pose and shape priors, with neural implicit functions that flexibly learn clothing details. However, this combined representation introduces additional computation, e.g. signed distance calculation in 3D body feature extraction, leading to redundancy in the implicit query-and-infer process and failing to preserve the underlying body shape prior. To address these issues, we propose a novel IUVD-Feedback representation, consisting of an IUVD occupancy function and a feedback query algorithm. This representation replaces the time-consuming signed distance calculation with a simple linear transformation in the IUVD space, leveraging the SMPL UV maps. Additionally, it reduces redundant query points through a feedback mechanism, leading to more reasonable 3D body features and more effective query points, thereby preserving the parametric body prior. Moreover, the IUVD-Feedback representation can be embedded into any existing implicit human reconstruction pipeline without requiring modifications to the trained neural networks. Experiments on the THuman2.0 dataset demonstrate that the proposed IUVD-Feedback representation improves the robustness of results and achieves three times faster acceleration in the query-and-infer process. Furthermore, this representation holds potential for generative applications by leveraging its inherent semantic information from the parametric body model.

Xu Zhao, Pengju Zhang, Bo Liu et al.

Monocular 3D occupancy prediction, aiming to predict the occupancy and semantics within interesting regions of 3D scenes from only 2D images, has garnered increasing attention recently for its vital role in 3D scene understanding. Predicting the 3D occupancy of large-scale outdoor scenes from 2D images is ill-posed and resource-intensive. In this paper, we present \textbf{DGOcc}, a \textbf{D}epth-aware \textbf{G}lobal query-based network for monocular 3D \textbf{Occ}upancy prediction. We first explore prior depth maps to extract depth context features that provide explicit geometric information for the occupancy network. Then, in order to fully exploit the depth context features, we propose a Global Query-based (GQ) Module. The cooperation of attention mechanisms and scale-aware operations facilitates the feature interaction between images and 3D voxels. Moreover, a Hierarchical Supervision Strategy (HSS) is designed to avoid upsampling the high-dimension 3D voxel features to full resolution, which mitigates GPU memory utilization and time cost. Extensive experiments on SemanticKITTI and SSCBench-KITTI-360 datasets demonstrate that the proposed method achieves the best performance on monocular semantic occupancy prediction while reducing GPU and time overhead.

Zhenyi Hou, Xu Zhao, Kejie Ye et al.

Vocal education in the music field is difficult to quantify due to the individual differences in singers' voices and the different quantitative criteria of singing techniques. Deep learning has great potential to be applied in music education due to its efficiency to handle complex data and perform quantitative analysis. However, accurate evaluations with limited samples over rare vocal types, such as Mezzo-soprano, requires extensive well-annotated data support using deep learning models. In order to attain the objective, we perform transfer learning by employing deep learning models pre-trained on the ImageNet and Urbansound8k datasets for the improvement on the precision of vocal technique evaluation. Furthermore, we tackle the problem of the lack of samples by constructing a dedicated dataset, the Mezzo-soprano Vocal Set (MVS), for vocal technique assessment. Our experimental results indicate that transfer learning increases the overall accuracy (OAcc) of all models by an average of 8.3%, with the highest accuracy at 94.2%. We not only provide a novel approach to evaluating Mezzo-soprano vocal techniques but also introduce a new quantitative assessment method for music education.

Xiangyu Shi, Junyang Ding, Xu Zhao et al.

Computer-aided design (CAD) is vital to modern manufacturing, yet model creation remains labor-intensive and expertise-heavy. To enable non-experts to translate intuitive design intent into manufacturable artifacts, recent large language models-based text-to-CAD efforts focus on command sequences or script-based formats like CadQuery. However, these formats are kernel-dependent and lack universality for manufacturing. In contrast, the Standard for the Exchange of Product Data (STEP, ISO 10303) file is a widely adopted, neutral boundary representation (B-rep) format directly compatible with manufacturing, but its graph-structured, cross-referenced nature poses unique challenges for auto-regressive LLMs. To address this, we curate a dataset of ~40K STEP-caption pairs and introduce novel preprocessing tailored for the graph-structured format of STEP, including a depth-first search-based reserialization that linearizes cross-references while preserving locality and chain-of-thought(CoT)-style structural annotations that guide global coherence. We integrate retrieval-augmented generation to ground predictions in relevant examples for supervised fine-tuning, and refine generation quality through reinforcement learning with a specific Chamfer Distance-based geometric reward. Experiments demonstrate consistent gains of our STEP-LLM in geometric fidelity over the Text2CAD baseline, with improvements arising from multiple stages of our framework: the RAG module substantially enhances completeness and renderability, the DFS-based reserialization strengthens overall accuracy, and the RL further reduces geometric discrepancy. Both metrics and visual comparisons confirm that STEP-LLM generates shapes with higher fidelity than Text2CAD. These results show the feasibility of LLM-driven STEP model generation from natural language, showing its potential to democratize CAD design for manufacturing.

Yong Deng, Baoxing Li, Xu Zhao

Reconstructing 3D clothed humans from monocular camera data is highly challenging due to viewpoint limitations and image ambiguity. While implicit function-based approaches, combined with prior knowledge from parametric models, have made significant progress, there are still two notable problems. Firstly, the back details of human models are ambiguous due to viewpoint invisibility. The quality of the back details depends on the back normal map predicted by a convolutional neural network (CNN). However, the CNN lacks global information awareness for comprehending the back texture, resulting in excessively smooth back details. Secondly, a single image suffers from local ambiguity due to lighting conditions and body movement. However, implicit functions are highly sensitive to pixel variations in ambiguous regions. To address these ambiguities, we propose the Spatial-Temporal Transformer (STT) network for 3D clothed human reconstruction. A spatial transformer is employed to extract global information for normal map prediction. The establishment of global correlations facilitates the network in comprehending the holistic texture and shape of the human body. Simultaneously, to compensate for local ambiguity in images, a temporal transformer is utilized to extract temporal features from adjacent frames. The incorporation of temporal features can enhance the accuracy of input features in implicit networks. Furthermore, to obtain more accurate temporal features, joint tokens are employed to establish local correspondences between frames. Experimental results on the Adobe and MonoPerfCap datasets have shown that our method outperforms state-of-the-art methods and maintains robust generalization even under low-light outdoor conditions.

Yiming Bao, Xu Zhao, Dahong Qian

Temporal 3D human pose estimation from monocular videos is a challenging task in human-centered computer vision due to the depth ambiguity of 2D-to-3D lifting. To improve accuracy and address occlusion issues, inertial sensor has been introduced to provide complementary source of information. However, it remains challenging to integrate heterogeneous sensor data for producing physically rational 3D human poses. In this paper, we propose a novel framework, Real-time Optimization and Fusion (RTOF), to address this issue. We first incorporate sparse inertial orientations into a parametric human skeleton to refine 3D poses in kinematics. The poses are then optimized by energy functions built on both visual and inertial observations to reduce the temporal jitters. Our framework outputs smooth and biomechanically plausible human motion. Comprehensive experiments with ablation studies demonstrate its rationality and efficiency. On Total Capture dataset, the pose estimation error is significantly decreased compared to the baseline method.

Yuxi Xie, Kenji Kawaguchi, Yiran Zhao et al.

Breaking down a problem into intermediate steps has demonstrated impressive performance in Large Language Model (LLM) reasoning. However, the growth of the reasoning chain introduces uncertainty and error accumulation, making it challenging to elicit accurate final results. To tackle this challenge of uncertainty in multi-step reasoning, we introduce a stepwise self-evaluation mechanism to guide and calibrate the reasoning process of LLMs. We propose a decoding algorithm integrating the self-evaluation guidance via stochastic beam search. The self-evaluation guidance serves as a better-calibrated automatic criterion, facilitating an efficient search in the reasoning space and resulting in superior prediction quality. Stochastic beam search balances exploitation and exploration of the search space with temperature-controlled randomness. Our approach surpasses the corresponding Codex-backboned baselines in few-shot accuracy by $6.34\%$, $9.56\%$, and $5.46\%$ on the GSM8K, AQuA, and StrategyQA benchmarks, respectively. Experiment results with Llama-2 on arithmetic reasoning demonstrate the efficiency of our method in outperforming the baseline methods with comparable computational budgets. Further analysis in multi-step reasoning finds our self-evaluation guidance pinpoints logic failures and leads to higher consistency and robustness. Our code is publicly available at https://guideddecoding.github.io/.

Yesheng Zhang, Xu Zhao, Dahong Qian

In this paper, we present a novel end-to-end network architecture to estimate fundamental matrix directly from stereo images. To establish a complete working pipeline, different deep neural networks in charge of finding correspondences in images, performing outlier rejection and calculating fundamental matrix, are integrated into an end-to-end network architecture. To well train the network and preserve geometry properties of fundamental matrix, a new loss function is introduced. To evaluate the accuracy of estimated fundamental matrix more reasonably, we design a new evaluation metric which is highly consistent with visualization result. Experiments conducted on both outdoor and indoor data-sets show that this network outperforms traditional methods as well as previous deep learning based methods on various metrics and achieves significant performance improvements.

Xu Zhao, Zihao Wang, Hao Wu et al.

Semi-supervision is a promising paradigm for Bilingual Lexicon Induction (BLI) with limited annotations. However, previous semisupervised methods do not fully utilize the knowledge hidden in annotated and nonannotated data, which hinders further improvement of their performance. In this paper, we propose a new semi-supervised BLI framework to encourage the interaction between the supervised signal and unsupervised alignment. We design two message-passing mechanisms to transfer knowledge between annotated and non-annotated data, named prior optimal transport and bi-directional lexicon update respectively. Then, we perform semi-supervised learning based on a cyclic or a parallel parameter feeding routine to update our models. Our framework is a general framework that can incorporate any supervised and unsupervised BLI methods based on optimal transport. Experimental results on MUSE and VecMap datasets show significant improvement of our models. Ablation study also proves that the two-way interaction between the supervised signal and unsupervised alignment accounts for the gain of the overall performance. Results on distant language pairs further illustrate the advantage and robustness of our proposed method.

Xu Zhao, Zihao Wang, Hao Wu et al.

Recently unsupervised Bilingual Lexicon Induction (BLI) without any parallel corpus has attracted much research interest. One of the crucial parts in methods for the BLI task is the matching procedure. Previous works impose a too strong constraint on the matching and lead to many counterintuitive translation pairings. Thus, We propose a relaxed matching procedure to find a more precise matching between two languages. We also find that aligning source and target language embedding space bidirectionally will bring significant improvement. We follow the previous iterative framework to conduct experiments. Results on standard benchmark demonstrate the effectiveness of our proposed method, which substantially outperforms previous unsupervised methods.