Yiheng Li

Yunxiang Zhang, Yiheng Li, Ali Payani et al.

A central challenge for language agents is utilizing past experience to adapt to dynamic test-time conditions. While recent work demonstrates the promise of agentic memory mechanisms, most systems restrict retrieval to episode initiation. Consequently, agents are forced to rely on static guidance that becomes increasingly misaligned as long-horizon tasks unfold. To address this rigidity, we propose the Adaptive Memory Agent (AdaMEM), a novel framework for agent test-time adaptation. Without updating model parameters online, AdaMEM adapts agent behavior via a hybrid memory architecture: it maintains a long-term trajectory memory of raw experiences collected offline while generating dynamic short-term strategy memory on-the-fly to guide decision-making. This mechanism enables the trade-off between token efficiency and adaptability across varying inference-time compute levels. Empirically, AdaMEM significantly outperforms static memory baselines, achieving relative gains of up to 13% on ALFWorld and 11% on WebShop, with consistent leading performance extending to agentic search on HotpotQA. To further enhance this adaptation, we develop STEP-MFT, a Step-wise Memory Fine-Tuning technique that trains the policy to synthesize high-quality strategies from retrieved experiences, yielding additional performance gains. Our work establishes a new scaling dimension for agentic memory, supporting continuous reasoning and self-evolution post-deployment in real-world environments. Our code is available at https://github.com/yunx-z/AdaMEM.

Yiheng Li, Seth Z. Zhao, Chenfeng Xu et al.

Accumulating substantial volumes of real-world driving data proves pivotal in the realm of trajectory forecasting for autonomous driving. Given the heavy reliance of current trajectory forecasting models on data-driven methodologies, we aim to tackle the challenge of learning general trajectory forecasting representations under limited data availability. We propose a pipeline-level solution to mitigate the issue of data scarcity in trajectory forecasting. The solution is composed of two parts: firstly, we adopt HD map augmentation and trajectory synthesis for generating driving data, and then we learn representations by pre-training on them. Specifically, we apply vector transformations to reshape the maps, and then employ a rule-based model to generate trajectories on both original and augmented scenes; thus enlarging the driving data without collecting additional real ones. To foster the learning of general representations within this augmented dataset, we comprehensively explore the different pre-training strategies, including extending the concept of a Masked AutoEncoder (MAE) for trajectory forecasting. Without bells and whistles, our proposed pipeline-level solution is general, simple, yet effective: we conduct extensive experiments to demonstrate the effectiveness of our data expansion and pre-training strategies, which outperform the baseline prediction model by large margins, e.g. 5.04%, 3.84% and 8.30% in terms of $MR_6$, $minADE_6$ and $minFDE_6$. The pre-training dataset and the codes for pre-training and fine-tuning are released at https://github.com/yhli123/Pretraining_on_Synthetic_Driving_Data_for_Trajectory_Prediction.

Yiheng Li, Zhuo Li, Ruibing Hou et al.

Conditional human motion generation remains a fundamental challenge in computer vision and robotics. Despite significant progress, current methods are often constrained by fixed modality configurations and task-specific architectures, leaving cross-modal interactions and the scaling laws of multimodal-conditioned synthesis largely underexplored. A key bottleneck is the scarcity of large-scale modality-aligned motion data, limiting generalization across diverse control signals. In this work, we introduce OmniHuMo, a large-scale, high-quality dataset comprising over 5,000 hours of motion and 3.2 million sequences with precisely aligned multimodal annotations (e.g., text, speech, music, and trajectory). Leveraging OmniHuMo, we propose AnyMo, a unified multimodal framework combining a Residual FSQ-based motion tokenizer with a scalable masked modeling transformer, enabling high-quality motion synthesis under arbitrary modality combinations. Extensive experiments show that AnyMo achieves high-fidelity synthesis while offering flexible control over both spatial and stylistic attributes.

Yiheng Li, Canhui Tang, Runzhao Yao et al.

Patch-to-point matching has become a robust way of point cloud registration. However, previous patch-matching methods employ superpoints with poor localization precision as nodes, which may lead to ambiguous patch partitions. In this paper, we propose a HybridPoint-based network to find more robust and accurate correspondences. Firstly, we propose to use salient points with prominent local features as nodes to increase patch repeatability, and introduce some uniformly distributed points to complete the point cloud, thus constituting hybrid points. Hybrid points not only have better localization precision but also give a complete picture of the whole point cloud. Furthermore, based on the characteristic of hybrid points, we propose a dual-classes patch matching module, which leverages the matching results of salient points and filters the matching noise of non-salient points. Experiments show that our model achieves state-of-the-art performance on 3DMatch, 3DLoMatch, and KITTI odometry, especially with 93.0% Registration Recall on the 3DMatch dataset. Our code and models are available at https://github.com/liyih/HybridPoint.

Yiming Liao, Yiheng Li, Ning Jiang et al.

Accurate computational prediction of T cell receptor (TCR) antigen specificity would transform the study of T cell biology and enable scalable immune engineering, yet existing models lack sufficient sensitivity and specificity for broad applications. A major limitation is the absence of rigorously defined, unseen benchmark datasets that allow unbiased evaluation of model performance and generalizability. Here, we describe two complementary classes of datasets that meet this criterion and argue that they provide both a robust framework for model assessment and a foundation for next-generation TCR-antigen prediction algorithm development.

Yizhi Chen, Zhanxiang Cao, Xinyi Peng et al.

Current Vision--Language--Action (VLA) models often optimize for semantic grounding, whereas executable manipulation requires geometry-aware spatial alignment and dynamic affordance selection. We introduce GeoAlign, a state-guided spatial alignment architecture for VLA policy learning. GeoAlign post-trains an RGB geometry branch with robot-domain RGB-D supervision, yielding RGB-derived Geometry-Enhanced Post-Trained (GEP) features for policy rollout. The robot's proprioceptive state queries the GEP feature grid, producing compact, phase-dependent geometry tokens for action prediction. GeoAlign achieves 99.0% on LIBERO, 85.3% across three SimplerEnv-Fractal tasks, and 78.8% on eight geometry-critical real-world ALOHA tasks, with ablations confirming the value of geometry post-training and proprioceptive-state-guided querying.

Yiheng Li, Connelly Barnes, Kun Huang et al.

Optical flow computation is essential in the early stages of the video processing pipeline. This paper focuses on a less explored problem in this area, the 360$^\circ$ optical flow estimation using deep neural networks to support increasingly popular VR applications. To address the distortions of panoramic representations when applying convolutional neural networks, we propose a novel multi-projection fusion framework that fuses the optical flow predicted by the models trained using different projection methods. It learns to combine the complementary information in the optical flow results under different projections. We also build the first large-scale panoramic optical flow dataset to support the training of neural networks and the evaluation of panoramic optical flow estimation methods. The experimental results on our dataset demonstrate that our method outperforms the existing methods and other alternative deep networks that were developed for processing 360° content.

Yiheng Li, Yang Yang, Zichang Tan et al.

As the misuse of AI-generated images grows, generalizable image detection techniques are urgently needed. Recent state-of-the-art (SOTA) methods adopt aligned training datasets to reduce content, size, and format biases, empowering models to capture robust forgery cues. A common strategy is to employ reconstruction techniques, e.g., VAE and DDIM, which show remarkable results in diffusion-based methods. However, such reconstruction-based approaches typically introduce limited and homogeneous artifacts, which cannot fully capture diverse generative patterns, such as GAN-based methods. To complement reconstruction-based fake images with aligned yet diverse artifact patterns, we propose a GAN-based upsampling approach that mimics GAN-generated fake patterns while preserving content, size, and format alignment. This naturally results in two aligned but distinct types of fake images. However, due to the domain shift between reconstruction-based and upsampling-based fake images, direct mixed training causes suboptimal results, where one domain disrupts feature learning of the other. Accordingly, we propose a Separate Expert Fusion (SEF) framework to extract complementary artifact information and reduce inter-domain interference. We first train domain-specific experts via LoRA adaptation on a frozen foundational model, then conduct decoupled fusion with a gating network to adaptively combine expert features while retaining their specialized knowledge. Rather than merely benefiting GAN-generated image detection, this design introduces diverse and complementary artifact patterns that enable SEF to learn a more robust decision boundary and improve generalization across broader generative methods. Extensive experiments demonstrate that our method yields strong results across 13 diverse benchmarks. Codes are released at: https://github.com/liyih/SEF_AIGC_detection.

Yiheng Li, Weihai Lu, Hanyi Yu et al.

In recent years, multimodal multidomain fake news detection has garnered increasing attention. Nevertheless, this direction presents two significant challenges: (1) Failure to Capture Cross-Instance Narrative Consistency: existing models usually evaluate each news in isolation, fail to capture cross-instance narrative consistency, and thus struggle to address the spread of cluster based fake news driven by social media; (2) Lack of Domain Specific Knowledge for Reasoning: conventional models, which rely solely on knowledge encoded in their parameters during training, struggle to generalize to new or data-scarce domains (e.g., emerging events or niche topics). To tackle these challenges, we introduce Retrieval-Augmented Multimodal Model for Fake News Detection (RAMM). First, RAMM employs a Multimodal Large Language Model (MLLM) as its backbone to capture cross-modal semantic information from news samples. Second, RAMM incorporates an Abstract Narrative Alignment Module. This component adaptively extracts abstract narrative consistency from diverse instances across distinct domains, aggregates relevant knowledge, and thereby enables the modeling of high-level narrative information. Finally, RAMM introduces a Semantic Representation Alignment Module, which aligns the model's decision-making paradigm with that of humans - specifically, it shifts the model's reasoning process from direct inference on multimodal features to an instance-based analogical reasoning process. Extensive experimental results on three public datasets validate the efficacy of our proposed approach. Our code is available at the following link: https://github.com/li-yiheng/RAMM

Xingyi Li, Jiahui Zhang, Yiheng Li et al.

As generative image models evolve rapidly, the perceptual gap between generated and real images continues to narrow, making AI-generated image detection increasingly challenging. Many existing methods exploit frequency-domain cues for detection, typically described as frequency-domain artifacts or high-frequency discrepancies. However, the specific and recurring spectral regularities remain insufficiently understood and characterized. In this paper, we systematically analyze the one-dimensional radial log-power spectra of real and generated images. We find that generated images do not necessarily exhibit higher or lower energy across the entire spectrum or high-band range. Instead, their spectra deviate from the power-law decay and show an anomalous uplift in the ultra-high-frequency tail. We term this phenomenon spectral tail uplift. We further attribute this phenomenon to nonlinear harmonic accumulation in trained generative models, suggesting that it can serve as a structural cue across generative architectures. Based on this observation, we propose Spectral Tail Auxiliary Learning (STAL), a frequency-domain auxiliary supervision framework for generalizable AI-generated image detection. STAL transfers spectral-tail cues from a tail-aware frequency teacher to a spatial detector during training, while all frequency-domain modules are discarded at inference time. Consequently, STAL introduces no inference overhead. Extensive experiments on 9 public datasets show that STAL achieves strong generalization and stability across generators, data distributions, and real-world scenarios.

Mingda Jia, Weiliang Meng, Zenghuang Fu et al.

Dense video captioning jointly localizes and captions salient events in untrimmed videos. Recent methods primarily focus on leveraging additional prior knowledge and advanced multi-task architectures to achieve competitive performance. However, these pipelines rely on implicit modeling that uses frame-level or fragmented video features, failing to capture the temporal coherence across event sequences and comprehensive semantics within visual contexts. To address this, we propose an explicit temporal-semantic modeling framework called Context-Aware Cross-Modal Interaction (CACMI), which leverages both latent temporal characteristics within videos and linguistic semantics from text corpus. Specifically, our model consists of two core components: Cross-modal Frame Aggregation aggregates relevant frames to extract temporally coherent, event-aligned textual features through cross-modal retrieval; and Context-aware Feature Enhancement utilizes query-guided attention to integrate visual dynamics with pseudo-event semantics. Extensive experiments on the ActivityNet Captions and YouCook2 datasets demonstrate that CACMI achieves the state-of-the-art performance on dense video captioning task.

Lei Zhu, Xing Cai, Yingjie Chen et al.

Recent advancements in audio-video joint generation models have demonstrated impressive capabilities in content creation. However, generating high-fidelity human-centric videos in complex, real-world physical scenes remains a significant challenge. We identify that the root cause lies in the structural deficiencies of existing datasets across three dimensions: limited global scene and camera diversity, sparse interaction modeling (both person-person and person-object), and insufficient individual attribute alignment. To bridge these gaps, we present OmniHuman, a large-scale, multi-scene dataset designed for fine-grained human modeling. OmniHuman provides a hierarchical annotation covering video-level scenes, frame-level interactions, and individual-level attributes. To facilitate this, we develop a fully automated pipeline for high-quality data collection and multi-modal annotation. Complementary to the dataset, we establish the OmniHuman Benchmark (OHBench), a three-level evaluation system that provides a scientific diagnosis for human-centric audio-video synthesis. Crucially, OHBench introduces metrics that are highly consistent with human perception, filling the gaps in existing benchmarks by providing a comprehensive diagnosis across global scenes, relational interactions, and individual attributes.

Yiheng Li, Yang Yang, Zhen Lei

In radar-camera 3D object detection, the radar point clouds are sparse and noisy, which causes difficulties in fusing camera and radar modalities. To solve this, we introduce a novel query-based detection method named Radar-Camera Transformer (RCTrans). Specifically, we first design a Radar Dense Encoder to enrich the sparse valid radar tokens, and then concatenate them with the image tokens. By doing this, we can fully explore the 3D information of each interest region and reduce the interference of empty tokens during the fusing stage. We then design a Pruning Sequential Decoder to predict 3D boxes based on the obtained tokens and random initialized queries. To alleviate the effect of elevation ambiguity in radar point clouds, we gradually locate the position of the object via a sequential fusion structure. It helps to get more precise and flexible correspondences between tokens and queries. A pruning training strategy is adopted in the decoder, which can save much time during inference and inhibit queries from losing their distinctiveness. Extensive experiments on the large-scale nuScenes dataset prove the superiority of our method, and we also achieve new state-of-the-art radar-camera 3D detection results. Our implementation is available at https://github.com/liyih/RCTrans.

Xiaoxi Cui, Weihai Lu, Yu Tong et al.

In click-through rate prediction, click-through rate prediction is used to model users' interests. However, most of the existing CTR prediction methods are mainly based on the ID modality. As a result, they are unable to comprehensively model users' multi-modal preferences. Therefore, it is necessary to introduce multi-modal CTR prediction. Although it seems appealing to directly apply the existing multi-modal fusion methods to click-through rate prediction models, these methods (1) fail to effectively disentangle commonalities and specificities across different modalities; (2) fail to consider the synergistic effects between modalities and model the complex interactions between modalities. To address the above issues, this paper proposes the Diffusion-based Multi-modal Synergy Interest Network (Diff-MSIN) framework for click-through prediction. This framework introduces three innovative modules: the Multi-modal Feature Enhancement (MFE) Module Synergistic Relationship Capture (SRC) Module, and the Feature Dynamic Adaptive Fusion (FDAF) Module. The MFE Module and SRC Module extract synergistic, common, and special information among different modalities. They effectively enhances the representation of the modalities, improving the overall quality of the fusion. To encourage distinctiveness among different features, we design a Knowledge Decoupling method. Additionally, the FDAF Module focuses on capturing user preferences and reducing fusion noise. To validate the effectiveness of the Diff-MSIN framework, we conducted extensive experiments using the Rec-Tmall and three Amazon datasets. The results demonstrate that our approach yields a significant improvement of at least 1.67% compared to the baseline, highlighting its potential for enhancing multi-modal recommendation systems. Our code is available at the following link: https://github.com/Cxx-0/Diff-MSIN.

Yiheng Li, Yang Yang, Zhen Lei

Fusing multi-modality inputs from different sensors is an effective way to improve the performance of 3D object detection. However, current methods overlook two important conflicts: point-pixel misalignment and sub-task suppression. The former means a pixel feature from the opaque object is projected to multiple point features of the same ray in the world space, and the latter means the classification prediction and bounding box regression may cause mutual suppression. In this paper, we propose a novel method named Conflict Resolution Network (CoreNet) to address the aforementioned issues. Specifically, we first propose a dual-stream transformation module to tackle point-pixel misalignment. It consists of ray-based and point-based 2D-to-BEV transformations. Both of them achieve approximately unique mapping from the image space to the world space. Moreover, we introduce a task-specific predictor to tackle sub-task suppression. It uses the dual-branch structure which adopts class-specific query and Bbox-specific query to corresponding sub-tasks. Each task-specific query is constructed of task-specific feature and general feature, which allows the heads to adaptively select information of interest based on different sub-tasks. Experiments on the large-scale nuScenes dataset demonstrate the superiority of our proposed CoreNet, by achieving 75.6\% NDS and 73.3\% mAP on the nuScenes test set without test-time augmentation and model ensemble techniques. The ample ablation study also demonstrates the effectiveness of each component. The code is released on https://github.com/liyih/CoreNet.

Qiyan Zhao, Xiaofeng Zhang, Yiheng Li et al.

Hallucinations pose a significant challenge in Large Vision Language Models (LVLMs), with misalignment between multimodal features identified as a key contributing factor. This paper reveals the negative impact of the long-term decay in Rotary Position Encoding (RoPE), used for positional modeling in LVLMs, on multimodal alignment. Concretely, under long-term decay, instruction tokens exhibit uneven perception of image tokens located at different positions within the two-dimensional space: prioritizing image tokens from the bottom-right region since in the one-dimensional sequence, these tokens are positionally closer to the instruction tokens. This biased perception leads to insufficient image-instruction interaction and suboptimal multimodal alignment. We refer to this phenomenon as image alignment bias. To enhance instruction's perception of image tokens at different spatial locations, we propose MCA-LLaVA, based on Manhattan distance, which extends the long-term decay to a two-dimensional, multi-directional spatial decay. MCA-LLaVA integrates the one-dimensional sequence order and two-dimensional spatial position of image tokens for positional modeling, mitigating hallucinations by alleviating image alignment bias. Experimental results of MCA-LLaVA across various hallucination and general benchmarks demonstrate its effectiveness and generality. The code can be accessed in https://github.com/ErikZ719/MCA-LLaVA.

Yiheng Li, Zichang Tan, Guoqing Xu et al.

With the rapid development of generative AI in medical imaging, synthetic Computed Tomography (CT) images have demonstrated great potential in applications such as data augmentation and clinical diagnosis, but they also introduce serious security risks. Despite the increasing security concerns, existing studies on CT forgery detection are still limited and fail to adequately address real-world challenges. These limitations are mainly reflected in two aspects: the absence of datasets that can effectively evaluate model generalization to reflect the real-world application requirements, and the reliance on detection methods designed for natural images that are insensitive to CT-specific forgery artifacts. In this view, we propose CTForensics, a comprehensive dataset designed to systematically evaluate the generalization capability of CT forgery detection methods, which includes ten diverse CT generative methods. Moreover, we introduce the Enhanced Spatial-Frequency CT Forgery Detector (ESF-CTFD), an efficient CNN-based neural network that captures forgery cues across the wavelet, spatial, and frequency domains. First, it transforms the input CT image into three scales and extracts features at each scale via the Wavelet-Enhanced Central Stem. Then, starting from the largest-scale features, the Spatial Process Block gradually performs feature fusion with the smaller-scale ones. Finally, the Frequency Process Block learns frequency-domain information for predicting the final results. Experiments demonstrate that ESF-CTFD consistently outperforms existing methods and exhibits superior generalization across different CT generative models.

Yiheng Li, Zichang Tan, Zhen Lei et al.

In AI-generated image detection, current cutting-edge methods typically adapt pre-trained foundation models through partial-parameter fine-tuning. However, these approaches often struggle to generalize to forgeries from unseen generators, as the fine-tuned models capture only limited patterns from training data and fail to reflect the evolving traits of new ones. To overcome this limitation, we propose Image-Adaptive Prompt Learning (IAPL), a novel paradigm that dynamically adjusts the prompts fed into the encoder according to each testing image, rather than fixing them after training. This design significantly enhances robustness and adaptability to diverse forged images. The dynamic prompts integrate conditional information with test-time adaptive tokens through a lightweight learnable scaling factor. The conditional information is produced by a Conditional Information Learner, which leverages CNN-based feature extractors to model both forgery-specific and general conditions. The test-time adaptive tokens are optimized during inference on a single sample by enforcing prediction consistency across multiple views, ensuring that the parameters align with the current image. For the final decision, the optimal input with the highest prediction confidence is selected. Extensive experiments show that IAPL achieves state-of-the-art performance, with mean accuracies of 95.61% and 96.7% on the widely used UniversalFakeDetect and GenImage datasets, respectively. Codes and weights will be released on https://github.com/liyih/IAPL.

Yiheng Li, Feng Liang, Dan Kondratyuk et al.

The substantial training cost of diffusion models hinders their deployment. Immiscible Diffusion recently showed that reducing diffusion trajectory mixing in the noise space via linear assignment accelerates training by simplifying denoising. To extend immiscible diffusion beyond the inefficient linear assignment under high batch sizes and high dimensions, we refine this concept to a broader miscibility reduction at any layer and by any implementation. Specifically, we empirically demonstrate the bijective nature of the denoising process with respect to immiscible diffusion, ensuring its preservation of generative diversity. Moreover, we provide thorough analysis and show step-by-step how immiscibility eases denoising and improves efficiency. Extending beyond linear assignment, we propose a family of implementations including K-nearest neighbor (KNN) noise selection and image scaling to reduce miscibility, achieving up to >4x faster training across diverse models and tasks including unconditional/conditional generation, image editing, and robotics planning. Furthermore, our analysis of immiscibility offers a novel perspective on how optimal transport (OT) enhances diffusion training. By identifying trajectory miscibility as a fundamental bottleneck, we believe this work establishes a potentially new direction for future research into high-efficiency diffusion training. The code is available at https://github.com/yhli123/Immiscible-Diffusion.

Yiheng Li, Heyang Jiang, Akio Kodaira et al.

In this paper, we point out that suboptimal noise-data mapping leads to slow training of diffusion models. During diffusion training, current methods diffuse each image across the entire noise space, resulting in a mixture of all images at every point in the noise layer. We emphasize that this random mixture of noise-data mapping complicates the optimization of the denoising function in diffusion models. Drawing inspiration from the immiscibility phenomenon in physics, we propose Immiscible Diffusion, a simple and effective method to improve the random mixture of noise-data mapping. In physics, miscibility can vary according to various intermolecular forces. Thus, immiscibility means that the mixing of molecular sources is distinguishable. Inspired by this concept, we propose an assignment-then-diffusion training strategy to achieve Immiscible Diffusion. As one example, prior to diffusing the image data into noise, we assign diffusion target noise for the image data by minimizing the total image-noise pair distance in a mini-batch. The assignment functions analogously to external forces to expel the diffuse-able areas of images, thus mitigating the inherent difficulties in diffusion training. Our approach is remarkably simple, requiring only one line of code to restrict the diffuse-able area for each image while preserving the Gaussian distribution of noise. In this way, each image is preferably projected to nearby noise. Experiments demonstrate that our method can achieve up to 3x faster training for unconditional Consistency Models on the CIFAR dataset, as well as for DDIM and Stable Diffusion on CelebA and ImageNet dataset, and in class-conditional training and fine-tuning. In addition, we conducted a thorough analysis that sheds light on how it improves diffusion training speed while improving fidelity. The code is available at https://yhli123.github.io/immiscible-diffusion

Yun Xing, Yiheng Li, Ivan Laptev et al.

Recent Large Vision Language Models (LVLMs) present remarkable zero-shot conversational and reasoning capabilities given multimodal queries. Nevertheless, they suffer from object hallucination, a phenomenon where LVLMs are prone to generate textual responses not factually aligned with image inputs. Our pilot study reveals that object hallucination is closely tied with Rotary Position Encoding (RoPE), a widely adopted positional dependency modeling design in existing LVLMs. Due to the long-term decay in RoPE, LVLMs tend to hallucinate more when relevant visual cues are distant from instruction tokens in the multimodal input sequence. Additionally, we observe a similar effect when reversing the sequential order of visual tokens during multimodal alignment. Our tests indicate that long-term decay in RoPE poses challenges to LVLMs while capturing visual-instruction interactions across long distances. We propose Concentric Causal Attention (CCA), a simple yet effective positional alignment strategy that mitigates the impact of RoPE long-term decay in LVLMs by naturally reducing relative distance between visual and instruction tokens. With CCA, visual tokens can better interact with instruction tokens, thereby enhancing model's perception capability and alleviating object hallucination. Without bells and whistles, our positional alignment method surpasses existing hallucination mitigation strategies by large margins on multiple object hallucination benchmarks.

Yiheng Li, Ruibing Hou, Hong Chang et al.

Human pose plays a crucial role in the digital age. While recent works have achieved impressive progress in understanding and generating human poses, they often support only a single modality of control signals and operate in isolation, limiting their application in real-world scenarios. This paper presents UniPose, a framework employing Large Language Models (LLMs) to comprehend, generate, and edit human poses across various modalities, including images, text, and 3D SMPL poses. Specifically, we apply a pose tokenizer to convert 3D poses into discrete pose tokens, enabling seamless integration into the LLM within a unified vocabulary. To further enhance the fine-grained pose perception capabilities, we facilitate UniPose with a mixture of visual encoders, among them a pose-specific visual encoder. Benefiting from a unified learning strategy, UniPose effectively transfers knowledge across different pose-relevant tasks, adapts to unseen tasks, and exhibits extended capabilities. This work serves as the first attempt at building a general-purpose framework for pose comprehension, generation, and editing. Extensive experiments highlight UniPose's competitive and even superior performance across various pose-relevant tasks.

Xiaoxi Cui, Weihai Lu, Yu Tong et al.

The sequential recommendation system utilizes historical user interactions to predict preferences. Effectively integrating diverse user behavior patterns with rich multimodal information of items to enhance the accuracy of sequential recommendations is an emerging and challenging research direction. This paper focuses on the problem of multi-modal multi-behavior sequential recommendation, aiming to address the following challenges: (1) the lack of effective characterization of modal preferences across different behaviors, as user attention to different item modalities varies depending on the behavior; (2) the difficulty of effectively mitigating implicit noise in user behavior, such as unintended actions like accidental clicks; (3) the inability to handle modality noise in multi-modal representations, which further impacts the accurate modeling of user preferences. To tackle these issues, we propose a novel Multi-Modal Multi-Behavior Sequential Recommendation model (M$^3$BSR). This model first removes noise in multi-modal representations using a Conditional Diffusion Modality Denoising Layer. Subsequently, it utilizes deep behavioral information to guide the denoising of shallow behavioral data, thereby alleviating the impact of noise in implicit feedback through Conditional Diffusion Behavior Denoising. Finally, by introducing a Multi-Expert Interest Extraction Layer, M$^3$BSR explicitly models the common and specific interests across behaviors and modalities to enhance recommendation performance. Experimental results indicate that M$^3$BSR significantly outperforms existing state-of-the-art methods on benchmark datasets.

Yiheng Li, Hongyang Li, Zehao Huang et al.

Multi-modal 3D object detection has exhibited significant progress in recent years. However, most existing methods can hardly scale to long-range scenarios due to their reliance on dense 3D features, which substantially escalate computational demands and memory usage. In this paper, we introduce SparseFusion, a novel multi-modal fusion framework fully built upon sparse 3D features to facilitate efficient long-range perception. The core of our method is the Sparse View Transformer module, which selectively lifts regions of interest in 2D image space into the unified 3D space. The proposed module introduces sparsity from both semantic and geometric aspects which only fill grids that foreground objects potentially reside in. Comprehensive experiments have verified the efficiency and effectiveness of our framework in long-range 3D perception. Remarkably, on the long-range Argoverse2 dataset, SparseFusion reduces memory footprint and accelerates the inference by about two times compared to dense detectors. It also achieves state-of-the-art performance with mAP of 41.2% and CDS of 32.1%. The versatility of SparseFusion is also validated in the temporal object detection task and 3D lane detection task. Codes will be released upon acceptance.

Pengcheng Wang, Qinghang Liu, Haotian Lin et al.

Learning domain adaptive policies that can generalize to unseen transition dynamics, remains a fundamental challenge in learning-based control. Substantial progress has been made through domain representation learning to capture domain-specific information, thus enabling domain-aware decision making. We analyze the process of learning domain representations through dynamical prediction and find that selecting contexts adjacent to the current step causes the learned representations to entangle static domain information with varying dynamical properties. Such mixture can confuse the conditioned policy, thereby constraining zero-shot adaptation. To tackle the challenge, we propose DADP (Domain Adaptive Diffusion Policy), which achieves robust adaptation through unsupervised disentanglement and domain-aware diffusion injection. First, we introduce Lagged Context Dynamical Prediction, a strategy that conditions future state estimation on a historical offset context; by increasing this temporal gap, we unsupervisedly disentangle static domain representations by filtering out transient properties. Second, we integrate the learned domain representations directly into the generative process by biasing the prior distribution and reformulating the diffusion target. Extensive experiments on challenging benchmarks across locomotion and manipulation demonstrate the superior performance, and the generalizability of DADP over prior methods. More visualization results are available on the https://outsider86.github.io/DomainAdaptiveDiffusionPolicy/.

Guoqing Xu, Yiheng Li, Yang Yang

Precise modeling of lane topology is essential for autonomous driving, as it directly impacts navigation and control decisions. Existing methods typically represent each lane with a single query and infer topological connectivity based on the similarity between lane queries. However, this kind of design struggles to accurately model complex lane structures, leading to unreliable topology prediction. In this view, we propose a Fine-Grained lane topology reasoning framework (TopoFG). It divides the procedure from bird's-eye-view (BEV) features to topology prediction via fine-grained queries into three phases, i.e., Hierarchical Prior Extractor (HPE), Region-Focused Decoder (RFD), and Robust Boundary-Point Topology Reasoning (RBTR). Specifically, HPE extracts global spatial priors from the BEV mask and local sequential priors from in-lane keypoint sequences to guide subsequent fine-grained query modeling. RFD constructs fine-grained queries by integrating the spatial and sequential priors. It then samples reference points in RoI regions of the mask and applies cross-attention with BEV features to refine the query representations of each lane. RBTR models lane connectivity based on boundary-point query features and further employs a topological denoising strategy to reduce matching ambiguity. By integrating spatial and sequential priors into fine-grained queries and applying a denoising strategy to boundary-point topology reasoning, our method precisely models complex lane structures and delivers trustworthy topology predictions. Extensive experiments on the OpenLane-V2 benchmark demonstrate that TopoFG achieves new state-of-the-art performance, with an OLS of 48.0 on subsetA and 45.4 on subsetB.

Nicholas Babey, Tiffany Gu, Yiheng Li et al.

For embodied agents to effectively understand and interact within the world around them, they require a nuanced comprehension of human actions grounded in physical space. Current action recognition models, often relying on RGB video, learn superficial correlations between patterns and action labels, so they struggle to capture underlying physical interaction dynamics and human poses in complex scenes. We propose a model architecture that grounds action recognition in physical space by fusing two powerful, complementary representations: V-JEPA 2's contextual, predictive world dynamics and CoMotion's explicit, occlusion-tolerant human pose data. Our model is validated on both the InHARD and UCF-19-Y-OCC benchmarks for general action recognition and high-occlusion action recognition, respectively. Our model outperforms three other baselines, especially within complex, occlusive scenes. Our findings emphasize a need for action recognition to be supported by spatial understanding instead of statistical pattern recognition.

Qinmei Xu, Yiheng Li, Xianghao Zhan et al.

Foundation models leveraging vision-language pretraining have shown promise in chest X-ray (CXR) interpretation, yet their real-world performance across diverse populations and diagnostic tasks remains insufficiently evaluated. This study benchmarks the diagnostic performance and generalizability of foundation models versus traditional convolutional neural networks (CNNs) on multinational CXR datasets. We evaluated eight CXR diagnostic models - five vision-language foundation models and three CNN-based architectures - across 37 standardized classification tasks using six public datasets from the USA, Spain, India, and Vietnam, and three private datasets from hospitals in China. Performance was assessed using AUROC, AUPRC, and other metrics across both shared and dataset-specific tasks. Foundation models outperformed CNNs in both accuracy and task coverage. MAVL, a model incorporating knowledge-enhanced prompts and structured supervision, achieved the highest performance on public (mean AUROC: 0.82; AUPRC: 0.32) and private (mean AUROC: 0.95; AUPRC: 0.89) datasets, ranking first in 14 of 37 public and 3 of 4 private tasks. All models showed reduced performance on pediatric cases, with average AUROC dropping from 0.88 +/- 0.18 in adults to 0.57 +/- 0.29 in children (p = 0.0202). These findings highlight the value of structured supervision and prompt design in radiologic AI and suggest future directions including geographic expansion and ensemble modeling for clinical deployment. Code for all evaluated models is available at https://drive.google.com/drive/folders/1B99yMQm7bB4h1sVMIBja0RfUu8gLktCE

Yiheng Li, Francisco Carrillo-Perez, Mohammed Alawad et al.

Lung cancer is the leading cause of cancer mortality worldwide, and non-invasive methods for detecting key mutations and staging are essential for improving patient outcomes. Here, we compare the performance of two machine learning models - FMCIB+XGBoost, a supervised model with domain-specific pretraining, and Dinov2+ABMIL, a self-supervised model with attention-based multiple-instance learning - on 3D lung nodule data from the Stanford Radiogenomics and Lung-CT-PT-Dx cohorts. In the task of KRAS and EGFR mutation detection, FMCIB+XGBoost consistently outperformed Dinov2+ABMIL, achieving accuracies of 0.846 and 0.883 for KRAS and EGFR mutations, respectively. In cancer staging, Dinov2+ABMIL demonstrated competitive generalization, achieving an accuracy of 0.797 for T-stage prediction in the Lung-CT-PT-Dx cohort, suggesting SSL's adaptability across diverse datasets. Our results emphasize the clinical utility of supervised models in mutation detection and highlight the potential of SSL to improve staging generalization, while identifying areas for enhancement in mutation sensitivity.

Canhui Tang, Yiheng Li, Shaoyi Du et al.

Feature Descriptors and Detectors are two main components of feature-based point cloud registration. However, little attention has been drawn to the explicit representation of local and global semantics in the learning of descriptors and detectors. In this paper, we present a framework that explicitly extracts dual-level descriptors and detectors and performs coarse-to-fine matching with them. First, to explicitly learn local and global semantics, we propose a hierarchical contrastive learning strategy, training the robust matching ability of high-level descriptors, and refining the local feature space using low-level descriptors. Furthermore, we propose to learn dual-level saliency maps that extract two groups of keypoints in two different senses. To overcome the weak supervision of binary matchability labels, we propose a ranking strategy to label the significance ranking of keypoints, and thus provide more fine-grained supervision signals. Finally, we propose a global-to-local matching scheme to obtain robust and accurate correspondences by leveraging the complementary dual-level features.Quantitative experiments on 3DMatch and KITTI odometry datasets show that our method achieves robust and accurate point cloud registration and outperforms recent keypoint-based methods.

Xianghao Zhan, Yiheng Li, Yuzhe Liu et al.

Traumatic brain injury can be caused by various types of head impacts. However, due to different kinematic characteristics, many brain injury risk estimation models are not generalizable across the variety of impacts that humans may sustain. The current definitions of head impact subtypes are based on impact sources (e.g., football, traffic accident), which may not reflect the intrinsic kinematic similarities of impacts across the impact sources. To investigate the potential new definitions of impact subtypes based on kinematics, 3,161 head impacts from various sources including simulation, college football, mixed martial arts, and car racing were collected. We applied the K-means clustering to cluster the impacts on 16 standardized temporal features from head rotation kinematics. Then, we developed subtype-specific ridge regression models for cumulative strain damage (using the threshold of 15%), which significantly improved the estimation accuracy compared with the baseline method which mixed impacts from different sources and developed one model (R^2 from 0.7 to 0.9). To investigate the effect of kinematic features, we presented the top three critical features (maximum resultant angular acceleration, maximum angular acceleration along the z-axis, maximum linear acceleration along the y-axis) based on regression accuracy and used logistic regression to find the critical points for each feature that partitioned the subtypes. This study enables researchers to define head impact subtypes in a data-driven manner, which leads to more generalizable brain injury risk estimation.

Xianghao Zhan, Yiheng Li, Yuzhe Liu et al.

Objective: Traumatic brain injury can be caused by head impacts, but many brain injury risk estimation models are not equally accurate across the variety of impacts that patients may undergo and the characteristics of different types of impacts are not well studied. We investigated the spectral characteristics of different head impact types with kinematics classification. Methods: Data was analyzed from 3,262 head impacts from lab reconstruction, American football, mixed martial arts, and publicly available car crash data. A random forest classifier with spectral densities of linear acceleration and angular velocity was built to classify head impact types (e.g., football, car crash, mixed martial arts). To test the classifier robustness, another 271 lab-reconstructed impacts were obtained from 5 other instrumented mouthguards. Finally, with the classifier, type-specific, nearest-neighbor regression models were built for brain strain. Results: The classifier reached a median accuracy of 96% over 1,000 random partitions of training and test sets. The most important features in the classification included both low-frequency and high-frequency features, both linear acceleration features and angular velocity features. Different head impact types had different distributions of spectral densities in low-frequency and high-frequency ranges (e.g., the spectral densities of MMA impacts were higher in high-frequency range than in the low-frequency range). The type-specific regression showed a generally higher R^2-value than baseline models without classification. Conclusion: The machine-learning-based classifier enables a better understanding of the impact kinematics spectral density in different sports, and it can be applied to evaluate the quality of impact-simulation systems and on-field data augmentation.

Xianghao Zhan, Yiheng Li, Yuzhe Liu et al.

Brain tissue deformation resulting from head impacts is primarily caused by rotation and can lead to traumatic brain injury. To quantify brain injury risk based on measurements of kinematics on the head, finite element (FE) models and various brain injury criteria based on different factors of these kinematics have been developed, but the contribution of different kinematic factors has not been comprehensively analyzed across different types of head impacts in a data-driven manner. To better design brain injury criteria, the predictive power of rotational kinematics factors, which are different in 1) the derivative order (angular velocity, angular acceleration, angular jerk), 2) the direction and 3) the power (e.g., square-rooted, squared, cubic) of the angular velocity, were analyzed based on different datasets including laboratory impacts, American football, mixed martial arts (MMA), NHTSA automobile crashworthiness tests and NASCAR crash events. Ordinary least squares regressions were built from kinematics factors to the 95\% maximum principal strain (MPS95), and we compared zero-order correlation coefficients, structure coefficients, commonality analysis, and dominance analysis. The angular acceleration, the magnitude, and the first power factors showed the highest predictive power for the majority of impacts including laboratory impacts, American football impacts, with few exceptions (angular velocity for MMA and NASCAR impacts). The predictive power of rotational kinematics in three directions (x: posterior-to-anterior, y: left-to-right, z: superior-to-inferior) of kinematics varied with different sports and types of head impacts.

Xianghao Zhan, Yiheng Li, Yuzhe Liu et al.

Multiple brain injury criteria (BIC) are developed to quickly quantify brain injury risks after head impacts. These BIC originated from different types of head impacts (e.g., sports and car crashes) are widely used in risk evaluation. However, the accuracy of using the BIC on brain injury risk estimation across different types of head impacts has not been evaluated. Physiologically, brain strain is often considered the key parameter of brain injury. To evaluate the BIC's risk estimation accuracy across five datasets comprising different head impact types, linear regression was used to model 95% maximum principal strain, 95% maximum principal strain at the corpus callosum, and cumulative strain damage (15%) on each of 18 BIC respectively. The results show a significant difference in the relationship between BIC and brain strain across datasets, indicating the same BIC value may suggest different brain strain in different head impact types. The accuracy of brain strain regression is generally decreasing if the BIC regression models are fit on a dataset with a different type of head impact rather than on the dataset with the same type. Given this finding, this study raises concerns for applying BIC to estimate the brain injury risks for head impacts different from the head impacts on which the BIC was developed.