arXiv subjects

Jianfei Yang

Publications and source records attributed to Jianfei Yang.

At least 91 records · Page 5Linked to original sources

arXiv2025

mmPred: Radar-based Human Motion Prediction in the Dark

Existing Human Motion Prediction (HMP) methods based on RGB-D cameras are sensitive to lighting conditions and raise privacy concerns, limiting their real-world applications such as firefighting and healthcare. Motivated by the robustness and privacy-preserving nature of millimeter-wave (mmWave) radar, this work introduces radar as a novel sensing modality for HMP, for the first time. Nevertheless, radar signals often suffer from specular reflections and multipath effects, resulting in noisy and temporally inconsistent measurements, such as body-part miss-detection. To address these radar-specific artifacts, we propose mmPred, the first diffusion-based framework tailored for radar-based HMP. mmPred introduces a dual-domain historical motion representation to guide the generation process, combining a Time-domain Pose Refinement (TPR) branch for learning fine-grained details and a Frequency-domain Dominant Motion (FDM) branch for capturing global motion trends and suppressing frame-level inconsistency. Furthermore, we design a Global Skeleton-relational Transformer (GST) as the diffusion backbone to model global inter-joint cooperation, enabling corrupted joints to dynamically aggregate information from others. Extensive experiments show that mmPred achieves state-of-the-art performance, outperforming existing methods by 8.6% on mmBody and 22% on mm-Fi.

cs.CV↗

arXiv2025

Benford's Curse: Tracing Digit Bias to Numerical Hallucination in LLMs

Large Language Models (LLMs) exhibit impressive performance on complex reasoning tasks, yet they frequently fail on basic numerical problems, producing incorrect outputs. Inspired by Benford's Law, a statistical pattern in which lower digits occur more frequently as leading digits, we hypothesize that the skewed digit distributions in web-collected corpora may be learned by LLMs during pretraining, leading to biased numerical generation. To investigate the hypothesis, we first examine whether digits frequencies in pretraining corpus (OLMo2) follows Benford's law. We then construct an evaluation benchmark in which the ground-truth digits are uniformly distributed within each of the seven numerical reasoning tasks. Our evaluation results demonstrate that leading open-source LLMs show a consistent pattern of digit bias that resembles Benford's law. Through logit-lens tracing and neuron-level dissection, we identify that this bias arises predominantly from a small subset of highly digit-selective feed-forward network (FFN) neurons in the deeper layers. Finally, we demonstrate that pruning these neurons mitigates imbalanced overgeneration and partially corrects erroneous outputs, providing causal evidence that fine-grained pretraining digit bias can propagate into model behavior. Our findings reveal a fundamental connection between corpus-level statistics and symbolic failure modes in LLMs, offering a new lens for diagnosing and mitigating hallucinations in numerical tasks.

cs.CL↗

arXiv2025

HiCoGen: Hierarchical Compositional Text-to-Image Generation in Diffusion Models via Reinforcement Learning

Recent advances in diffusion models have demonstrated impressive capability in generating high-quality images for simple prompts. However, when confronted with complex prompts involving multiple objects and hierarchical structures, existing models struggle to accurately follow instructions, leading to issues such as concept omission, confusion, and poor compositionality. To address these limitations, we propose a Hierarchical Compositional Generative framework (HiCoGen) built upon a novel Chain of Synthesis (CoS) paradigm. Instead of monolithic generation, HiCoGen first leverages a Large Language Model (LLM) to decompose complex prompts into minimal semantic units. It then synthesizes these units iteratively, where the image generated in each step provides crucial visual context for the next, ensuring all textual concepts are faithfully constructed into the final scene. To further optimize this process, we introduce a reinforcement learning (RL) framework. Crucially, we identify that the limited exploration of standard diffusion samplers hinders effective RL. We theoretically prove that sample diversity is maximized by concentrating stochasticity in the early generation stages and, based on this insight, propose a novel Decaying Stochasticity Schedule to enhance exploration. Our RL algorithm is then guided by a hierarchical reward mechanism that jointly evaluates the image at the global, subject, and relationship levels. We also construct HiCoPrompt, a new text-to-image benchmark with hierarchical prompts for rigorous evaluation. Experiments show our approach significantly outperforms existing methods in both concept coverage and compositional accuracy.

cs.CV↗

arXiv2025

Mask2IV: Interaction-Centric Video Generation via Mask Trajectories

Generating interaction-centric videos, such as those depicting humans or robots interacting with objects, is crucial for embodied intelligence, as they provide rich and diverse visual priors for robot learning, manipulation policy training, and affordance reasoning. However, existing methods often struggle to model such complex and dynamic interactions. While recent studies show that masks can serve as effective control signals and enhance generation quality, obtaining dense and precise mask annotations remains a major challenge for real-world use. To overcome this limitation, we introduce Mask2IV, a novel framework specifically designed for interaction-centric video generation. It adopts a decoupled two-stage pipeline that first predicts plausible motion trajectories for both actor and object, then generates a video conditioned on these trajectories. This design eliminates the need for dense mask inputs from users while preserving the flexibility to manipulate the interaction process. Furthermore, Mask2IV supports versatile and intuitive control, allowing users to specify the target object of interaction and guide the motion trajectory through action descriptions or spatial position cues. To support systematic training and evaluation, we curate two benchmarks covering diverse action and object categories across both human-object interaction and robotic manipulation scenarios. Extensive experiments demonstrate that our method achieves superior visual realism and controllability compared to existing baselines.

cs.CV↗

arXiv2025

Zero-Shot Open-Vocabulary Human Motion Grounding with Test-Time Training

Understanding complex human activities demands the ability to decompose motion into fine-grained, semantic-aligned sub-actions. This motion grounding process is crucial for behavior analysis, embodied AI and virtual reality. Yet, most existing methods rely on dense supervision with predefined action classes, which are infeasible in open-vocabulary, real-world settings. In this paper, we propose ZOMG, a zero-shot, open-vocabulary framework that segments motion sequences into semantically meaningful sub-actions without requiring any annotations or fine-tuning. Technically, ZOMG integrates (1) language semantic partition, which leverages large language models to decompose instructions into ordered sub-action units, and (2) soft masking optimization, which learns instance-specific temporal masks to focus on frames critical to sub-actions, while maintaining intra-segment continuity and enforcing inter-segment separation, all without altering the pretrained encoder. Experiments on three motion-language datasets demonstrate state-of-the-art effectiveness and efficiency of motion grounding performance, outperforming prior methods by +8.7\% mAP on HumanML3D benchmark. Meanwhile, significant improvements also exist in downstream retrieval, establishing a new paradigm for annotation-free motion understanding.

cs.CV↗

arXiv2025

SignalLLM: A General-Purpose LLM Agent Framework for Automated Signal Processing

Modern signal processing (SP) pipelines, whether model-based or data-driven, often constrained by complex and fragmented workflow, rely heavily on expert knowledge and manual engineering, and struggle with adaptability and generalization under limited data. In contrast, Large Language Models (LLMs) offer strong reasoning capabilities, broad general-purpose knowledge, in-context learning, and cross-modal transfer abilities, positioning them as powerful tools for automating and generalizing SP workflows. Motivated by these potentials, we introduce SignalLLM, the first general-purpose LLM-based agent framework for general SP tasks. Unlike prior LLM-based SP approaches that are limited to narrow applications or tricky prompting, SignalLLM introduces a principled, modular architecture. It decomposes high-level SP goals into structured subtasks via in-context learning and domain-specific retrieval, followed by hierarchical planning through adaptive retrieval-augmented generation (RAG) and refinement; these subtasks are then executed through prompt-based reasoning, cross-modal reasoning, code synthesis, model invocation, or data-driven LLM-assisted modeling. Its generalizable design enables the flexible selection of problem solving strategies across different signal modalities, task types, and data conditions. We demonstrate the versatility and effectiveness of SignalLLM through five representative tasks in communication and sensing, such as radar target detection, human activity recognition, and text compression. Experimental results show superior performance over traditional and existing LLM-based methods, particularly in few-shot and zero-shot settings.

cs.LG↗

arXiv2025

ResearchGPT: Benchmarking and Training LLMs for End-to-End Computer Science Research Workflows

As large language models (LLMs) advance, the ultimate vision for their role in science is emerging: we could build an AI collaborator to effectively assist human beings throughout the entire scientific research process. We refer to this envisioned system as ResearchGPT. Given that scientific research progresses through multiple interdependent phases, achieving this vision requires rigorous benchmarks that evaluate the end-to-end workflow rather than isolated sub-tasks. To this end, we contribute CS-54k, a high-quality corpus of scientific Q&A pairs in computer science, built from 14k CC-licensed papers. It is constructed through a scalable, paper-grounded pipeline that combines retrieval-augmented generation (RAG) with multi-stage quality control to ensure factual grounding. From this unified corpus, we derive two complementary subsets: CS-4k, a carefully curated benchmark for evaluating AI's ability to assist scientific research, and CS-50k, a large-scale training dataset. Extensive experiments demonstrate that CS-4k stratifies state-of-the-art LLMs into distinct capability tiers. Open models trained on CS-50k with supervised training and reinforcement learning demonstrate substantial improvements. Even 7B-scale models, when properly trained, outperform many larger proprietary systems, such as GPT-4.1, GPT-4o, and Gemini 2.5 Pro. This indicates that making AI models better research assistants relies more on domain-aligned training with high-quality data than on pretraining scale or general benchmark performance. We release CS-4k and CS-50k in the hope of fostering AI systems as reliable collaborators in CS research.

cs.LG↗

arXiv2025

Interactive Test-Time Adaptation with Reliable Spatial-Temporal Voxels for Multi-Modal Segmentation

Multi-modal test-time adaptation (MM-TTA) adapts models to an unlabeled target domain by leveraging the complementary multi-modal inputs in an online manner. While previous MM-TTA methods for 3D segmentation offer a promising solution by leveraging self-refinement per frame, they suffer from two major limitations: 1) unstable frame-wise predictions caused by temporal inconsistency, and 2) consistently incorrect predictions that violate the assumption of reliable modality guidance. To address these limitations, this work introduces a comprehensive two-fold framework. Firstly, building upon our previous work ReLiable Spatial-temporal Voxels (Latte), we propose Latte++ that better suppresses the unstable frame-wise predictions with more informative geometric correspondences. Instead of utilizing a universal sliding window, Latte++ employs multi-window aggregation to capture more reliable correspondences to better evaluate the local prediction consistency of different semantic categories. Secondly, to tackle the consistently incorrect predictions, we propose Interactive Test-Time Adaptation (ITTA), a flexible add-on to empower effortless human feedback with existing MM-TTA methods. ITTA introduces a novel human-in-the-loop approach that efficiently integrates minimal human feedback through interactive segmentation, requiring only simple point clicks and bounding box annotations. Instead of using independent interactive networks, ITTA employs a lightweight promptable branch with a momentum gradient module to capture and reuse knowledge from scarce human feedback during online inference. Extensive experiments across five MM-TTA benchmarks demonstrate that ITTA achieves consistent and notable improvements with robust performance gains for target classes of interest in challenging imbalanced scenarios, while Latte++ provides complementary benefits for temporal stability.

cs.CV↗

arXiv2025

AV-PedAware: Self-Supervised Audio-Visual Fusion for Dynamic Pedestrian Awareness

In this study, we introduce AV-PedAware, a self-supervised audio-visual fusion system designed to improve dynamic pedestrian awareness for robotics applications. Pedestrian awareness is a critical requirement in many robotics applications. However, traditional approaches that rely on cameras and LIDARs to cover multiple views can be expensive and susceptible to issues such as changes in illumination, occlusion, and weather conditions. Our proposed solution replicates human perception for 3D pedestrian detection using low-cost audio and visual fusion. This study represents the first attempt to employ audio-visual fusion to monitor footstep sounds for the purpose of predicting the movements of pedestrians in the vicinity. The system is trained through self-supervised learning based on LIDAR-generated labels, making it a cost-effective alternative to LIDAR-based pedestrian awareness. AV-PedAware achieves comparable results to LIDAR-based systems at a fraction of the cost. By utilizing an attention mechanism, it can handle dynamic lighting and occlusions, overcoming the limitations of traditional LIDAR and camera-based systems. To evaluate our approach's effectiveness, we collected a new multimodal pedestrian detection dataset and conducted experiments that demonstrate the system's ability to provide reliable 3D detection results using only audio and visual data, even in extreme visual conditions. We will make our collected dataset and source code available online for the community to encourage further development in the field of robotics perception systems.

cs.RO↗

arXiv2025

Generative Dataset Distillation using Min-Max Diffusion Model

In this paper, we address the problem of generative dataset distillation that utilizes generative models to synthesize images. The generator may produce any number of images under a preserved evaluation time. In this work, we leverage the popular diffusion model as the generator to compute a surrogate dataset, boosted by a min-max loss to control the dataset's diversity and representativeness during training. However, the diffusion model is time-consuming when generating images, as it requires an iterative generation process. We observe a critical trade-off between the number of image samples and the image quality controlled by the diffusion steps and propose Diffusion Step Reduction to achieve optimal performance. This paper details our comprehensive method and its performance. Our model achieved $2^{nd}$ place in the generative track of \href{https://www.dd-challenge.com/#/}{The First Dataset Distillation Challenge of ECCV2024}, demonstrating its superior performance.

cs.CV↗

arXiv2025

RAW-Adapter: Adapting Pre-trained Visual Model to Camera RAW Images and A Benchmark

In the computer vision community, the preference for pre-training visual models has largely shifted toward sRGB images due to their ease of acquisition and compact storage. However, camera RAW images preserve abundant physical details across diverse real-world scenarios. Despite this, most existing visual perception methods that utilize RAW data directly integrate image signal processing (ISP) stages with subsequent network modules, often overlooking potential synergies at the model level. Building on recent advances in adapter-based methodologies in both NLP and computer vision, we propose RAW-Adapter, a novel framework that incorporates learnable ISP modules as input-level adapters to adjust RAW inputs. At the same time, it employs model-level adapters to seamlessly bridge ISP processing with high-level downstream architectures. Moreover, RAW-Adapter serves as a general framework applicable to various computer vision frameworks. Furthermore, we introduce RAW-Bench, which incorporates 17 types of RAW-based common corruptions, including lightness degradations, weather effects, blurriness, camera imaging degradations, and variations in camera color response. Using this benchmark, we systematically compare the performance of RAW-Adapter with state-of-the-art (SOTA) ISP methods and other RAW-based high-level vision algorithms. Additionally, we propose a RAW-based data augmentation strategy to further enhance RAW-Adapter's performance and improve its out-of-domain (OOD) generalization ability. Extensive experiments substantiate the effectiveness and efficiency of RAW-Adapter, highlighting its robust performance across diverse scenarios.

cs.CV↗

arXiv2025

QLIO: Quantized LiDAR-Inertial Odometry

LiDAR-Inertial Odometry (LIO) is widely used for autonomous navigation, but its deployment on Size, Weight, and Power (SWaP)-constrained platforms remains challenging due to the computational cost of processing dense point clouds. Conventional LIO frameworks rely on a single onboard processor, leading to computational bottlenecks and high memory demands, making real-time execution difficult on embedded systems. To address this, we propose QLIO, a multi-processor distributed quantized LIO framework that reduces computational load and bandwidth consumption while maintaining localization accuracy. QLIO introduces a quantized state estimation pipeline, where a co-processor pre-processes LiDAR measurements, compressing point-to-plane residuals before transmitting only essential features to the host processor. Additionally, an rQ-vector-based adaptive resampling strategy intelligently selects and compresses key observations, further reducing computational redundancy. Real-world evaluations demonstrate that QLIO achieves a 14.1% reduction in per-observation residual data while preserving localization accuracy. Furthermore, we release an open-source implementation to facilitate further research and real-world deployment. These results establish QLIO as an efficient and scalable solution for real-time autonomous systems operating under computational and bandwidth constraints.

cs.RO↗

arXiv2025

Feedback Favors the Generalization of Neural ODEs

The well-known generalization problem hinders the application of artificial neural networks in continuous-time prediction tasks with varying latent dynamics. In sharp contrast, biological systems can neatly adapt to evolving environments benefiting from real-time feedback mechanisms. Inspired by the feedback philosophy, we present feedback neural networks, showing that a feedback loop can flexibly correct the learned latent dynamics of neural ordinary differential equations (neural ODEs), leading to a prominent generalization improvement. The feedback neural network is a novel two-DOF neural network, which possesses robust performance in unseen scenarios with no loss of accuracy performance on previous tasks.} A linear feedback form is presented to correct the learned latent dynamics firstly, with a convergence guarantee. Then, domain randomization is utilized to learn a nonlinear neural feedback form. Finally, extensive tests including trajectory prediction of a real irregular object and model predictive control of a quadrotor with various uncertainties, are implemented, indicating significant improvements over state-of-the-art model-based and learning-based methods.

cs.LG↗

arXiv2025

Identifying and Mitigating Social Bias Knowledge in Language Models

Generating fair and accurate predictions plays a pivotal role in deploying large language models (LLMs) in the real world. However, existing debiasing methods inevitably generate unfair or incorrect predictions as they are designed and evaluated to achieve parity across different social groups but leave aside individual commonsense facts, resulting in modified knowledge that elicits unreasonable or undesired predictions. In this paper, we first establish a new bias mitigation benchmark, BiaScope, which systematically assesses performance by leveraging newly constructed datasets and metrics on knowledge retention and generalization. Then, we propose a novel debiasing approach, Fairness Stamp (FAST), which enables fine-grained calibration of individual social biases. FAST identifies the decisive layer responsible for storing social biases and then calibrates its outputs by integrating a small modular network, considering both bias mitigation and knowledge-preserving demands. Comprehensive experiments demonstrate that FAST surpasses state-of-the-art baselines with superior debiasing performance while not compromising the overall model capability for knowledge retention and downstream predictions. This highlights the potential of fine-grained debiasing strategies to achieve fairness in LLMs.

cs.CL↗

arXiv2025

X-Fi: A Modality-Invariant Foundation Model for Multimodal Human Sensing

Human sensing, which employs various sensors and advanced deep learning technologies to accurately capture and interpret human body information, has significantly impacted fields like public security and robotics. However, current human sensing primarily depends on modalities such as cameras and LiDAR, each of which has its own strengths and limitations. Furthermore, existing multi-modal fusion solutions are typically designed for fixed modality combinations, requiring extensive retraining when modalities are added or removed for diverse scenarios. In this paper, we propose a modality-invariant foundation model for all modalities, X-Fi, to address this issue. X-Fi enables the independent or combinatory use of sensor modalities without additional training by utilizing a transformer structure to accommodate variable input sizes and incorporating a novel "X-fusion" mechanism to preserve modality-specific features during multimodal integration. This approach not only enhances adaptability but also facilitates the learning of complementary features across modalities. Extensive experiments conducted on the MM-Fi and XRF55 datasets, employing six distinct modalities, demonstrate that X-Fi achieves state-of-the-art performance in human pose estimation (HPE) and human activity recognition (HAR) tasks. The findings indicate that our proposed model can efficiently support a wide range of human sensing applications, ultimately contributing to the evolution of scalable, multimodal sensing technologies.

cs.CV↗

arXiv2025

Audio Array-Based 3D UAV Trajectory Estimation with LiDAR Pseudo-Labeling

As small unmanned aerial vehicles (UAVs) become increasingly prevalent, there is growing concern regarding their impact on public safety and privacy, highlighting the need for advanced tracking and trajectory estimation solutions. In response, this paper introduces a novel framework that utilizes audio array for 3D UAV trajectory estimation. Our approach incorporates a self-supervised learning model, starting with the conversion of audio data into mel-spectrograms, which are analyzed through an encoder to extract crucial temporal and spectral information. Simultaneously, UAV trajectories are estimated using LiDAR point clouds via unsupervised methods. These LiDAR-based estimations act as pseudo labels, enabling the training of an Audio Perception Network without requiring labeled data. In this architecture, the LiDAR-based system operates as the Teacher Network, guiding the Audio Perception Network, which serves as the Student Network. Once trained, the model can independently predict 3D trajectories using only audio signals, with no need for LiDAR data or external ground truth during deployment. To further enhance precision, we apply Gaussian Process modeling for improved spatiotemporal tracking. Our method delivers top-tier performance on the MMAUD dataset, establishing a new benchmark in trajectory estimation using self-supervised learning techniques without reliance on ground truth annotations.

cs.RO↗

arXiv2025

Unsupervised UAV 3D Trajectories Estimation with Sparse Point Clouds

Compact UAV systems, while advancing delivery and surveillance, pose significant security challenges due to their small size, which hinders detection by traditional methods. This paper presents a cost-effective, unsupervised UAV detection method using spatial-temporal sequence processing to fuse multiple LiDAR scans for accurate UAV tracking in real-world scenarios. Our approach segments point clouds into foreground and background, analyzes spatial-temporal data, and employs a scoring mechanism to enhance detection accuracy. Tested on a public dataset, our solution placed 4th in the CVPR 2024 UG2+ Challenge, demonstrating its practical effectiveness. We plan to open-source all designs, code, and sample data for the research community github.com/lianghanfang/UnLiDAR-UAV-Est.

cs.CV↗

arXiv2025

Emergence of Painting Ability via Recognition-Driven Evolution

From Paleolithic cave paintings to Impressionism, human painting has evolved to depict increasingly complex and detailed scenes, conveying more nuanced messages. This paper attempts to emerge this artistic capability by simulating the evolutionary pressures that enhance visual communication efficiency. Specifically, we present a model with a stroke branch and a palette branch that together simulate human-like painting. The palette branch learns a limited colour palette, while the stroke branch parameterises each stroke using Bézier curves to render an image, subsequently evaluated by a high-level recognition module. We quantify the efficiency of visual communication by measuring the recognition accuracy achieved with machine vision. The model then optimises the control points and colour choices for each stroke to maximise recognition accuracy with minimal strokes and colours. Experimental results show that our model achieves superior performance in high-level recognition tasks, delivering artistic expression and aesthetic appeal, especially in abstract sketches. Additionally, our approach shows promise as an efficient bit-level image compression technique, outperforming traditional methods.

cs.CV↗