arXiv subjects

Zhiyuan Zhang

Publications and source records attributed to Zhiyuan Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2025

FlexTraj: Image-to-Video Generation with Flexible Point Trajectory Control

We present FlexTraj, a framework for image-to-video generation with flexible point trajectory control. FlexTraj introduces a unified point-based motion representation that encodes each point with a segmentation ID, a temporally consistent trajectory ID, and an optional color channel for appearance cues, enabling both dense and sparse trajectory control. Instead of injecting trajectory conditions into the video generator through token concatenation or ControlNet, FlexTraj employs an efficient sequence-concatenation scheme that achieves faster convergence, stronger controllability, and more efficient inference, while maintaining robustness under unaligned conditions. To train such a unified point trajectory-controlled video generator, FlexTraj adopts an annealing training strategy that gradually reduces reliance on complete supervision and aligned condition. Experimental results demonstrate that FlexTraj enables multi-granularity, alignment-agnostic trajectory control for video generation, supporting various applications such as motion cloning, drag-based image-to-video, motion interpolation, camera redirection, flexible action control and mesh animations.

cs.CV↗

arXiv2025

MoGIC: Boosting Motion Generation via Intention Understanding and Visual Context

Existing text-driven motion generation methods often treat synthesis as a bidirectional mapping between language and motion, but remain limited in capturing the causal logic of action execution and the human intentions that drive behavior. The absence of visual grounding further restricts precision and personalization, as language alone cannot specify fine-grained spatiotemporal details. We propose MoGIC, a unified framework that integrates intention modeling and visual priors into multimodal motion synthesis. By jointly optimizing multimodal-conditioned motion generation and intention prediction, MoGIC uncovers latent human goals, leverages visual priors to enhance generation, and exhibits versatile multimodal generative capability. We further introduce a mixture-of-attention mechanism with adaptive scope to enable effective local alignment between conditional tokens and motion subsequences. To support this paradigm, we curate Mo440H, a 440-hour benchmark from 21 high-quality motion datasets. Experiments show that after finetuning, MoGIC reduces FID by 38.6\% on HumanML3D and 34.6\% on Mo440H, surpasses LLM-based methods in motion captioning with a lightweight text head, and further enables intention prediction and vision-conditioned generation, advancing controllable motion synthesis and intention understanding. The code is available at https://github.com/JunyuShi02/MoGIC

cs.CV↗

arXiv2025

Efficient Active Training for Deep LiDAR Odometry

Robust and efficient deep LiDAR odometry models are crucial for accurate localization and 3D reconstruction, but typically require extensive and diverse training data to adapt to diverse environments, leading to inefficiencies. To tackle this, we introduce an active training framework designed to selectively extract training data from diverse environments, thereby reducing the training load and enhancing model generalization. Our framework is based on two key strategies: Initial Training Set Selection (ITSS) and Active Incremental Selection (AIS). ITSS begins by breaking down motion sequences from general weather into nodes and edges for detailed trajectory analysis, prioritizing diverse sequences to form a rich initial training dataset for training the base model. For complex sequences that are difficult to analyze, especially under challenging snowy weather conditions, AIS uses scene reconstruction and prediction inconsistency to iteratively select training samples, refining the model to handle a wide range of real-world scenarios. Experiments across datasets and weather conditions validate our approach's effectiveness. Notably, our method matches the performance of full-dataset training with just 52\% of the sequence volume, demonstrating the training efficiency and robustness of our active training paradigm. By optimizing the training process, our approach sets the stage for more agile and reliable LiDAR odometry systems, capable of navigating diverse environmental conditions with greater precision.

cs.RO↗

arXiv2025

Generalizing Unsupervised Lidar Odometry Model from Normal to Snowy Weather Conditions

Deep learning-based LiDAR odometry is crucial for autonomous driving and robotic navigation, yet its performance under adverse weather, especially snowfall, remains challenging. Existing models struggle to generalize across conditions due to sensitivity to snow-induced noise, limiting real-world use. In this work, we present an unsupervised LiDAR odometry model to close the gap between clear and snowy weather conditions. Our approach focuses on effective denoising to mitigate the impact of snowflake noise and outlier points on pose estimation, while also maintaining computational efficiency for real-time applications. To achieve this, we introduce a Patch Spatial Measure (PSM) module that evaluates the dispersion of points within each patch, enabling effective detection of sparse and discrete noise. We further propose a Patch Point Weight Predictor (PPWP) to assign adaptive point-wise weights, enhancing their discriminative capacity within local regions. To support real-time performance, we first apply an intensity threshold mask to quickly suppress dense snowflake clusters near the LiDAR, and then perform multi-modal feature fusion to refine the point-wise weight prediction, improving overall robustness under adverse weather. Our model is trained in clear weather conditions and rigorously tested across various scenarios, including snowy and dynamic. Extensive experimental results confirm the effectiveness of our method, demonstrating robust performance in both clear and snowy weather. This advancement enhances the model's generalizability and paves the way for more reliable autonomous systems capable of operating across a wider range of environmental conditions.

cs.RO↗

arXiv2025

A non-invasive dry-transfer method for fabricating mesoscopic devices on sensitive materials

Many materials with novel or exotic properties are highly sensitive to environmental factors such as air, solvents, and heat, which complicates device fabrication and limits their potential applications. Here, we present a universal submicron fabrication method for mesoscopic devices using a dry-transfer technique, tailored specifically for sensitive materials. This approach utilizes PMMA masks, combined with a water-dissoluble coating as a sacrificial layer, to ensure that sensitive materials are processed without exposure to harmful environmental conditions. The entire fabrication process is carried out in a glove box, employing dry techniques that avoid air, solvents, and heat exposure, culminating in an encapsulation step. We demonstrate the utility of this method by fabricating and characterizing K2Cr3As3 and WTe2 devices, a one- and two-dimensional material, respectively. The results show that our technique preserves the integrity of the materials, provides excellent contact interfaces, and is broadly applicable to a range of sensitive materials.

cond-mat.mes-hall↗

arXiv2025

Spectral radius and rainbow $k$-factors of graphs

Let $\mathcal{G}=\{G_1,\ldots, G_{\frac{kn}{2}}\}$ be a set of graphs on the same vertex set $V=\{1,\dots,n\}$ where $k\cdot n$ is even. We say $\mathcal{G}$ admits a rainbow $k$-factor if there exists a $k$-regular graph $F$ on the vertex set $V$ such that all edges of $F$ are from different members of $\mathcal{G}$. In this paper, we show a sufficient spectral condition for the existence of a rainbow $k$-factor for $k\geq 2$, which is that if $ρ(G_i)\geqρ(K_{k-1}\vee(K_1\cup K_{n-k}))$ for each $G_i\in \mathcal{G}$, then $\mathcal{G}$ admits a rainbow $k$-factor unless $G_1=G_2=\cdots=G_{\frac{kn}{2}}\cong K_{k-1}\vee(K_1\cup K_{n-k})$.

math.CO↗

arXiv2025

Density of States (Gate) - Controlled Andreev Molecule and Sensor

Topological quantum computing typically relies on topological Andreev bound states (ABSs) engineered in hybrid superconductor-semiconductor devices, where gate control offers key advantages. While strong Zeeman fields can induce such states, an alternative approach emerges through Andreev molecules -- closely spaced, coupled ABSs, also key building-block for Kitaev chain -- that enable topological behavior without high magnetic fields. However, existing Andreev molecules are controlled via magnetic flux in superconducting loops, limiting scalability. Here, we introduce a gate-controlled Andreev molecule, where electrostatic tuning of the density of states in one site nonlocally enhances the critical current of another. This eliminates superconducting loops, offering superior tunability, scalability, and sensitivity. We further extend such an Andreev molecule to a multi-site Kitaev chain, and a noninvasive sensor resolving single-Cooper-pair charge for parity readout. This platform bridges the gap between scalable ABS engineering and high-sensitivity quantum sensing, advancing the development for constructing and parity-readout in topological ABSs and long Kitaev chains towards topological qubits.

cond-mat.mes-hall↗

arXiv2025

LSFDNet: A Single-Stage Fusion and Detection Network for Ships Using SWIR and LWIR

Traditional ship detection methods primarily rely on single-modal approaches, such as visible or infrared images, which limit their application in complex scenarios involving varying lighting conditions and heavy fog. To address this issue, we explore the advantages of short-wave infrared (SWIR) and long-wave infrared (LWIR) in ship detection and propose a novel single-stage image fusion detection algorithm called LSFDNet. This algorithm leverages feature interaction between the image fusion and object detection subtask networks, achieving remarkable detection performance and generating visually impressive fused images. To further improve the saliency of objects in the fused images and improve the performance of the downstream detection task, we introduce the Multi-Level Cross-Fusion (MLCF) module. This module combines object-sensitive fused features from the detection task and aggregates features across multiple modalities, scales, and tasks to obtain more semantically rich fused features. Moreover, we utilize the position prior from the detection task in the Object Enhancement (OE) loss function, further increasing the retention of object semantics in the fused images. The detection task also utilizes preliminary fused features from the fusion task to complement SWIR and LWIR features, thereby enhancing detection performance. Additionally, we have established a Nearshore Ship Long-Short Wave Registration (NSLSR) dataset to train effective SWIR and LWIR image fusion and detection networks, bridging a gap in this field. We validated the superiority of our proposed single-stage fusion detection algorithm on two datasets. The source code and dataset are available at https://github.com/Yanyin-Guo/LSFDNet

cs.CV↗

arXiv2025

GenM$^3$: Generative Pretrained Multi-path Motion Model for Text Conditional Human Motion Generation

Scaling up motion datasets is crucial to enhance motion generation capabilities. However, training on large-scale multi-source datasets introduces data heterogeneity challenges due to variations in motion content. To address this, we propose Generative Pretrained Multi-path Motion Model (GenM$^3$), a comprehensive framework designed to learn unified motion representations. GenM$^3$ comprises two components: 1) a Multi-Expert VQ-VAE (MEVQ-VAE) that adapts to different dataset distributions to learn a unified discrete motion representation, and 2) a Multi-path Motion Transformer (MMT) that improves intra-modal representations by using separate modality-specific pathways, each with densely activated experts to accommodate variations within that modality, and improves inter-modal alignment by the text-motion shared pathway. To enable large-scale training, we integrate and unify 11 high-quality motion datasets (approximately 220 hours of motion data) and augment it with textual annotations (nearly 10,000 motion sequences labeled by a large language model and 300+ by human experts). After training on our integrated dataset, GenM$^3$ achieves a state-of-the-art FID of 0.035 on the HumanML3D benchmark, surpassing state-of-the-art methods by a large margin. It also demonstrates strong zero-shot generalization on IDEA400 dataset, highlighting its effectiveness and adaptability across diverse motion scenarios.

cs.CV↗

arXiv2025

Information-Bottleneck Driven Binary Neural Network for Change Detection

In this paper, we propose Binarized Change Detection (BiCD), the first binary neural network (BNN) designed specifically for change detection. Conventional network binarization approaches, which directly quantize both weights and activations in change detection models, severely limit the network's ability to represent input data and distinguish between changed and unchanged regions. This results in significantly lower detection accuracy compared to real-valued networks. To overcome these challenges, BiCD enhances both the representational power and feature separability of BNNs, improving detection performance. Specifically, we introduce an auxiliary objective based on the Information Bottleneck (IB) principle, guiding the encoder to retain essential input information while promoting better feature discrimination. Since directly computing mutual information under the IB principle is intractable, we design a compact, learnable auxiliary module as an approximation target, leading to a simple yet effective optimization strategy that minimizes both reconstruction loss and standard change detection loss. Extensive experiments on street-view and remote sensing datasets demonstrate that BiCD establishes a new benchmark for BNN-based change detection, achieving state-of-the-art performance in this domain.

cs.CV↗

arXiv2025

High-Fidelity Differential-information Driven Binary Vision Transformer

The binarization of vision transformers (ViTs) offers a promising approach to addressing the trade-off between high computational/storage demands and the constraints of edge-device deployment. However, existing binary ViT methods often suffer from severe performance degradation or rely heavily on full-precision modules. To address these issues, we propose DIDB-ViT, a novel binary ViT that is highly informative while maintaining the original ViT architecture and computational efficiency. Specifically, we design an informative attention module incorporating differential information to mitigate information loss caused by binarization and enhance high-frequency retention. To preserve the fidelity of the similarity calculations between binary Q and K tensors, we apply frequency decomposition using the discrete Haar wavelet and integrate similarities across different frequencies. Additionally, we introduce an improved RPReLU activation function to restructure the activation distribution, expanding the model's representational capacity. Experimental results demonstrate that our DIDB-ViT significantly outperforms state-of-the-art network quantization methods in multiple ViT architectures, achieving superior image classification and segmentation performance.

cs.CV↗

arXiv2025

DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving

End-to-end autonomous driving (E2E-AD) has emerged as a trend in the field of autonomous driving, promising a data-driven, scalable approach to system design. However, existing E2E-AD methods usually adopt the sequential paradigm of perception-prediction-planning, which leads to cumulative errors and training instability. The manual ordering of tasks also limits the system`s ability to leverage synergies between tasks (for example, planning-aware perception and game-theoretic interactive prediction and planning). Moreover, the dense BEV representation adopted by existing methods brings computational challenges for long-range perception and long-term temporal fusion. To address these challenges, we present DriveTransformer, a simplified E2E-AD framework for the ease of scaling up, characterized by three key features: Task Parallelism (All agent, map, and planning queries direct interact with each other at each block), Sparse Representation (Task queries direct interact with raw sensor features), and Streaming Processing (Task queries are stored and passed as history information). As a result, the new framework is composed of three unified operations: task self-attention, sensor cross-attention, temporal cross-attention, which significantly reduces the complexity of system and leads to better training stability. DriveTransformer achieves state-of-the-art performance in both simulated closed-loop benchmark Bench2Drive and real world open-loop benchmark nuScenes with high FPS.

cs.LG↗

arXiv2025

TrajTok: Technical Report for 2025 Waymo Open Sim Agents Challenge

In this technical report, we introduce TrajTok, a trajectory tokenizer for discrete next-token-prediction based behavior generation models, which combines data-driven and rule-based methods with better coverage, symmetry and robustness, along with a spatial-aware label smoothing method for cross-entropy loss. We adopt the tokenizer and loss for the SMART model and reach a superior performance with realism score of 0.7852 on the Waymo Open Sim Agents Challenge 2025. We will open-source the code in the future.

cs.CL↗

arXiv2025

ControlVLA: Few-shot Object-centric Adaptation for Pre-trained Vision-Language-Action Models

Learning real-world robotic manipulation is challenging, particularly when limited demonstrations are available. Existing methods for few-shot manipulation often rely on simulation-augmented data or pre-built modules like grasping and pose estimation, which struggle with sim-to-real gaps and lack extensibility. While large-scale imitation pre-training shows promise, adapting these general-purpose policies to specific tasks in data-scarce settings remains unexplored. To achieve this, we propose ControlVLA, a novel framework that bridges pre-trained VLA models with object-centric representations via a ControlNet-style architecture for efficient fine-tuning. Specifically, to introduce object-centric conditions without overwriting prior knowledge, ControlVLA zero-initializes a set of projection layers, allowing them to gradually adapt the pre-trained manipulation policies. In real-world experiments across 6 diverse tasks, including pouring cubes and folding clothes, our method achieves a 76.7% success rate while requiring only 10-20 demonstrations -- a significant improvement over traditional approaches that require more than 100 demonstrations to achieve comparable success. Additional experiments highlight ControlVLA's extensibility to long-horizon tasks and robustness to unseen objects and backgrounds.

cs.RO↗

arXiv2025

Time-Lapse Video-Based Embryo Grading via Complementary Spatial-Temporal Pattern Mining

Artificial intelligence has recently shown promise in automated embryo selection for In-Vitro Fertilization (IVF). However, current approaches either address partial embryo evaluation lacking holistic quality assessment or target clinical outcomes inevitably confounded by extra-embryonic factors, both limiting clinical utility. To bridge this gap, we propose a new task called Video-Based Embryo Grading - the first paradigm that directly utilizes full-length time-lapse monitoring (TLM) videos to predict embryologists' overall quality assessments. To support this task, we curate a real-world clinical dataset comprising over 2,500 TLM videos, each annotated with a grading label indicating the overall quality of embryos. Grounded in clinical decision-making principles, we propose a Complementary Spatial-Temporal Pattern Mining (CoSTeM) framework that conceptually replicates embryologists' evaluation process. The CoSTeM comprises two branches: (1) a morphological branch using a Mixture of Cross-Attentive Experts layer and a Temporal Selection Block to select discriminative local structural features, and (2) a morphokinetic branch employing a Temporal Transformer to model global developmental trajectories, synergistically integrating static and dynamic determinants for grading embryos. Extensive experimental results demonstrate the superiority of our design. This work provides a valuable methodological framework for AI-assisted embryo selection. The dataset and source code will be publicly available upon acceptance.

cs.CV↗

arXiv2025

Enhanced and modulable induced superconducting gap and effective Landé g-factor in Pb-InSb hybrid devices

The hybrid system of a conventional superconductor (SC) on a semiconductor (SM) nanowire with strong spin-orbit coupling (SOC) represents a promising platform for achieving topological superconductivity and Majorana zero modes (MZMs) towards topological quantum computation. While aluminum (Al)-based hybrid nanowire devices have been widely utilized, their limited superconducting gap and intrinsic weak SOC as well as small Landé g-factor may hinder future experimental advancements. In contrast, we demonstrate that lead (Pb)-based hybrid quantum devices exhibit a remarkably large and hard proximity-induced superconducting gap, exceeding that of Al by an order of magnitude. By exploiting electrostatic gating to modulate wavefunction distribution and SC-SM interfacial coupling, this gap can be continuously tuned from its maximum value (~1.4 meV, matching the bulk Pb gap) down to nearly zero while maintaining the hardness. Furthermore, magnetic-field-dependent measurements reveal a radial evolution of the gap structure with anti-crossing feature, indicative of strong SOC and huge effective g-factors up to 76. These findings underscore the superior functionality of Pb-based hybrid systems, significantly advancing their potential for realizing and stabilizing MZMs and the further scalable topological quantum architectures.

cond-mat.mes-hall↗

arXiv2025

ExoGait-MS: Learning Periodic Dynamics with Multi-Scale Graph Network for Exoskeleton Gait Recognition

Current exoskeleton control methods often face challenges in delivering personalized treatment. Standardized walking gaits can lead to patient discomfort or even injury. Therefore, personalized gait is essential for the effectiveness of exoskeleton robots, as it directly impacts their adaptability, comfort, and rehabilitation outcomes for individual users. To enable personalized treatment in exoskeleton-assisted therapy and related applications, accurate recognition of personal gait is crucial for implementing tailored gait control. The key challenge in gait recognition lies in effectively capturing individual differences in subtle gait features caused by joint synergy, such as step frequency and step length. To tackle this issue, we propose a novel approach, which uses Multi-Scale Global Dense Graph Convolutional Networks (GCN) in the spatial domain to identify latent joint synergy patterns. Moreover, we propose a Gait Non-linear Periodic Dynamics Learning module to effectively capture the periodic characteristics of gait in the temporal domain. To support our individual gait recognition task, we have constructed a comprehensive gait dataset that ensures both completeness and reliability. Our experimental results demonstrate that our method achieves an impressive accuracy of 94.34% on this dataset, surpassing the current state-of-the-art (SOTA) by 3.77%. This advancement underscores the potential of our approach to enhance personalized gait control in exoskeleton-assisted therapy.

cs.RO↗

arXiv2025

Circuit-level-configurable Zero-field Superconducting Diodes: A Universal Platform Beyond Intrinsic Symmetry Breaking

Modern industry seeks next-generation microelectronics with ultra-low dissipation and noise beyond semiconducting systems, where the superconducting electronics offer promise. Its physical foundation is the superconducting diode effect (SDE) with nonreciprocal supercurrent. SDE has hitherto mainly relied on material-specific intrinsic symmetry breaking in superconductors, suffering from low yield, controllability, and compatibility with further functional extension - an undesirable aspect for applications. Here, we demonstrated a field-free SDE due to the chemical potential shift from external circuit line resistance, which is generic and challenges the previous interpretations of the intrinsic symmetry breaking in superconductivity for zero-field SDE. Moreover, this SDE is circuit-level configurable since it can be electrically switched on/off with its polarity and efficiency precisely modulated via gate voltage and circuit reconfiguration, facilitating functional extension. Such a generic, controllable and extensible SDE addresses critical challenges in dissipationless circuit towards application, and thus establishes a robust platform for scalable superconducting electronics.

cond-mat.supr-con↗