CJCaigui Jiang

Caigui Jiang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Caigui Jiang has a total of 102 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • InteractiveAvatar: Real-Time Streaming Video Generation for Consistent and Intent-Aware Avatars

    by Quanyue Song, Yishan He, Yanfei Zhang, Shihao Cheng, Zhixiang He, Zhizhi Guo, Chi Zhang, Xuelong Li, Caigui Jiang


  • Open Access

    Unified framework for photon dynamics in non-Euclidean polygonal cavities

    by Yechun Ding, Yongsheng Wang, Peng Li, Yaxin Guo, Yanpeng Zhang, Feng Yun, Ce Shang, Run-Qiu Yang, Caigui Jiang, Feng Li

    We extend the concept of polygonal resonators to curved spaces through a unified model governed by a joint curvature parameter. This framework uncovers unique dynamics, including dissipative states defined by hyperbolic fixed points that yield phase diagrams in the curvature space. We find that stability transitions across geometric boundaries are characterized by sharp declines in the quality factor, a manifestation of chaotic instability that is overcome in the wave regime, highlighting the wave nature of wave chaos. A key discovery is the symmetry-driven asymptotic behavior in geodesic-sided cavities, revealing a profound connection between physical dynamics and underlying geometry. We also introduce a “non-Euclidean unit circle” as a general shape-invariant platform for optical simulation. These findings provide a robust framework for studying non-Euclidean wave chaos and point to strategies for designing integrated photonics with enhanced degrees of freedom.

  • Unlocking Euclidean problems with isotropic initialization

    by Khusrav Yorov, Bolun Wang, Mikhail Skopenkov, Helmut Pottmann, Caigui Jiang


  • Cascaded Robust Rectification for Arbitrary Document Images

    by Chaoyun Wang, Quanxin Huang, I-Chao Shen, Takeo Igarashi, Nanning Zheng, Caigui Jiang

    Document rectification in real-world scenarios poses significant challenges due to extreme variations in camera perspectives and physical distortions. Driven by the insight that complex transformations can be decomposed and resolved progressively, we introduce a novel multi-stage framework that progressively reverses distinct distortion types in a coarse-to-fine manner. Specifically, our framework first performs a global affine transformation to correct perspective distortions arising from the camera's viewpoint, then rectifies geometric deformations resulting from physical paper curling and folding, and finally employs a content-aware iterative process to eliminate fine-grained content distortions. To address limitations in existing evaluation protocols, we also propose two enhanced metrics: layout-aligned OCR metrics (AED/ACER) for a stable assessment that decouples geometric rectification quality from the layout analysis errors of OCR engines, and masked AD/AAD (AD-M/AAD-M) tailored for accurately evaluating geometric distortions in documents with incomplete boundaries. Extensive experiments show that our method establishes new state-of-the-art performance on multiple challenging benchmarks, yielding a substantial reduction of 14.1%–34.7% in the AAD metric and demonstrating superior efficacy in real-world applications.

  • Open Access

    SyncDIT: audio-visual aligned video generation with audio synchronization feature

    by Quanyue Song, Zhizhi Guo, Yishan He, Zhihao Wang, Zhixiang He, Chi Zhang, Caigui Jiang, Xuelong Li

    Audio-driven video generation has recently shown significant progress in producing realistic visual content. However, most previous studies focus primarily on speech or other narrow-band audio sources, limiting their effectiveness in scenarios involving complex auditory inputs, such as musical performances that combine transient sounds with sustained tones. To overcome these constraints, we introduce SyncDiT, a novel framework designed for synchronized audio-visual generation from complex audio signals and reference images. At its core, SyncDiT leverages an audio encoder to capture intricate acoustic characteristics within an audio clip, while an Audio Sync Cross-Attention module aligns video latents with frame-level audio synchronization features. Additionally, we construct the Musical Instrument Video (MIV) dataset, comprising diverse high-resolution instrument performance videos to support this task. Extensive experiments demonstrate that SyncDiT achieves superior audio-visual synchronization and enhanced visual quality compared to existing approaches, underscoring its effectiveness in generating lifelike videos from complex audio inputs.


  • Open Access

    H-ViT: hardware-friendly post-training quantization for efficient vision transformer inference

    by Jing Liu, Jiaqi Lai, Xiaodong Deng, Caigui Jiang, Nanning Zheng

    Vision Transformers (ViTs) have achieved state-of-the-art performance on various computer vision tasks. However these models are memory-consuming and computation-intensive, making their deployment and efficient inference on edge devices challenging. Model quantization is a promising approach to reduce model complexity. Prior works have explored tailored quantization algorithms for ViTs but unfortunately retained floating-point (FP) scaling factors, which not only yield non-negligible re-quantization overhead, but also hinder the quantized models to perform efficient integer-only inference. In this paper, we propose H-ViT, a dedicated post-training quantization scheme (e.g., symmetric uniform quantization and layer-wise quantization for both weights and part of activations) to effectively quantize ViTs with fewer Power-of-Two (PoT) scaling factors, thus minimizing the re-quantization overhead and memory consumption. In addition, observing serious inter-channel variation in LayerNorm inputs and outputs, we propose Power-of-Two quantization (PTQ), a systematic method to reducing the performance degradation without hyper-parameters. Extensive experiments are conducted on multiple vision tasks with different model variants, proving that H-ViT offers comparable(or even slightly higher) INT8 quantization performance with PoT scaling factors when compared to the counterpart with floating-point scaling factors. For instance, we reach 78.43 top-1 accuracy with DeiT-S on ImageNet, 51.6 box AP and 44.8 mask AP with Cascade Mask R-CNN (Swin-B) on COCO.


  • A Survey of Architectural Geometry for Design and Fabrication

    by Jianlei Wang, Zifeng Yu, Hui Wang, Caigui Jiang

    面向设计与制造的建筑几何综述 王建磊 1,2,3) , 于子峰 1,2,3) , 王慧 4) , 蒋才桂 1,2,3)* 1) (人机混合增强智能全国重点实验室 西安 710049) 2) (视觉信息与应用国家工程研究中心 西安 710049) 3) (西安交通大学人工智能与机器人研究所 西安 710049) 4) (西安交通大学数学与统计学院 西安 710049)

  • Open Access

    MCGM: Multi-stage Clustered Global Modeling for Long-range Interactions in Molecules

    by Haodong Pan, Yusong Wang, Nanning Zheng, Caijui Jiang

    Geometric graph neural networks (GNNs) excel at capturing molecular geometry, yet their locality-biased message passing hampers the modeling of long-range interactions. Current solutions have fundamental limitations: extending cutoff radii causes computational costs to scale cubically with distance; physics-inspired kernels (e.g., Coulomb, dispersion) are often system-specific and lack generality; Fourier-space methods require careful tuning of multiple parameters (e.g., mesh size, k-space cutoff) with added computational overhead. We introduce Multi-stage Clustered Global Modeling (MCGM), a lightweight, plug-and-play module that endows geometric GNNs with hierarchical global context through efficient clustering operations. MCGM builds a multi-resolution hierarchy of atomic clusters, distills global information via dynamic hierarchical clustering, and propagates this context back through learned transformations, ultimately reinforcing atomic features via residual connections. Seamlessly integrated into four diverse backbone architectures, MCGM reduces OE62 energy prediction error by an average of 26.2%. On AQM, MCGM achieves state-of-the-art accuracy (17.0 meV for energy, 4.9 meV/Å for forces) while using 20% fewer parameters than Neural P3M. Code will be made available upon acceptance.

  • Interactive design of developable surfaces by patch-based learning

    by Chaoyun Wang, Jianlei Wang, Chengcheng Tang, Nanning Zheng, Caigui Jiang