arXiv subjects

Jiajun Zhao

Publications and source records attributed to Jiajun Zhao.

7 recordsLinked to original sources

arXiv2026

Avatar V: Scaling Video-Reference Avatar Video Generation

Generating avatar videos that are not merely visually similar to a target individual but behaviorally recognizable, faithfully reproducing their talking rhythm, gestural tendencies, and expression dynamics, remains an open challenge. Existing methods predominantly condition on single static images, which provide insufficient identity information and cannot capture dynamic motion traits, while standard pixel-level objectives underserve the perceptually critical facial regions that determine avatar fidelity. We present Avatar V, a production-scale framework that addresses these limitations through video-reference-conditioned identity modeling. Rather than compressing identity into fixed-size embeddings, the model conditions directly on the full token sequence of a reference video, learning to reproduce both static identity attributes (facial geometry, skin texture) and dynamic behavioral patterns (talking rhythm, micro-expressions) through attention over the reference context. We introduce Sparse Reference Attention, an asymmetric mechanism achieving linear-complexity conditioning on arbitrarily long references; a motion representation stream enabling closed-loop talking style transfer; and an identity-aware super-resolution refiner inheriting the full reference conditioning. These are supported by a data engine curating 100M+ training clips from 50M raw videos, and a five-stage training pipeline with flow matching pre-training, personality fine-tuning, two-phase distillation (>10x acceleration), and RLHF alignment, deployed across thousands of GPUs. Avatar V generates 1080p videos of unlimited duration, achieving state-of-the-art identity preservation, lip synchronization, and generation quality on our cross-scene benchmark, consistently outperforming leading systems including Seedance 2.0, Kling O3 Pro, Veo 3.1, and OmniHuman 1.5 in both automated metrics and human evaluation.

cs.CV↗

arXiv2026

Intra-Class Subdivision for Pixel Contrastive Learning: Application to Semi-supervised Cardiac Image Segmentation

We propose an intra-class subdivision pixel contrastive learning (SPCL) framework for cardiac image segmentation to address representation contamination at boundaries. The novel concept ``Unconcerned sample'' is proposed to distinguish pixel representations at the inner and boundary regions within the same class, facilitating a clearer characterization of intra-class variations. A novel boundary contrastive loss for boundary representations is proposed to enhance representation discrimination across boundaries. The advantages of the unconcerned sample and boundary contrastive loss are analyzed theoretically. Experimental results in public cardiac datasets demonstrate that SPCL significantly improves segmentation performance, outperforming existing methods with respect to segmentation quality and boundary precision. Our code is available at https://github.com/Jrstud203/SPCL.

cs.CV↗

arXiv2023

A broadband hyperspectral image sensor with high spatio-temporal resolution

Hyperspectral imaging provides high-dimensional spatial-temporal-spectral information revealing intrinsic matter characteristics. Here we report an on-chip computational hyperspectral imaging framework with high spatial and temporal resolution. By integrating different broadband modulation materials on the image sensor chip, the target spectral information is non-uniformly and intrinsically coupled on each pixel with high light throughput. Using intelligent reconstruction algorithms, multi-channel images can be recovered from each frame, realizing real-time hyperspectral imaging. Following such a framework, we for the first time fabricated a broadband VIS-NIR (400-1700 nm) hyperspectral imaging sensor using photolithography, with an average light throughput of 74.8% and 96 wavelength channels. The demonstrated resolution is 1,024*1,024 pixels at 124 fps. We demonstrated its wide applications including chlorophyll and sugar quantification for intelligent agriculture, blood oxygen and water quality monitoring for human health, textile classification and apple bruise detection for industrial automation, and remote lunar detection for astronomy. The integrated hyperspectral image sensor weighs only tens of grams, and can be assembled on various resource-limited platforms or equipped with off-the-shelf optical systems. The technique transforms the challenge of high-dimensional imaging from a high-cost manufacturing and cumbersome system to one that is solvable through on-chip compression and agile computation.

physics.optics↗

arXiv2020

All-angle Negative Reflection with An Ultrathin Acoustic Gradient Metasurface: Floquet-Bloch Modes Perspective and Experimental Verification

Metasurface with gradient phase response offers new alternative for steering the propagation of waves. Conventional Snell's law has been revised by taking the contribution of local phase gradient into account. However, the requirement of momentum matching along the metasurface sets its nontrivial beam manipulation functionality within a limited-angle incidence. In this work, we theoretically and experimentally demonstrate that the acoustic gradient metasurface supports the negative reflection for all-angle incidence. The mode expansion theory is developed to help understand how the gradient metasurface tailors the incident beams, and the all-angle negative reflection occurs when the first negative order Floquet-Bloch mode dominates inside the metasurface slab. The coiling-up space structures are utilized to build desired acoustic gradient metasurface, and the all-angle negative reflections have been perfectly verified by experimental measurements. Our work offers the Floquet-Bloch modes perspective for qualitatively understanding the reflection behaviors of the acoustic gradient metasurface, and the all-angle negative reflection characteristic possessed by acoustic gradient metasurface could enable a new degree of the acoustic wave manipulating and be applied in the functional diffractive acoustic elements, such as the all-angle acoustic back reflector.

physics.class-ph↗

arXiv2017

Acoustic frequency filter based on anisotropic topological phononic crystals

There are growing efforts in constructing topological edge states in classical wave system. However, most of the work study the existence, creation and properties of the edge states, and the demonstration of application is highly desirable. Here, we present our design of a two-dimensional anisotropic phononic crystal that exhibits tunable topological phases. We further explore the contribution of anisotropy and show that the bandgap topology is also related to particular directions and frequency. Such frequency dependent behavior can be utilized as a frequency filter.

physics.app-ph↗

arXiv2014

Robust acoustic cloaking with density-near-zero materials

Using the phenomenon of extraordinary sound transmission (EST) through artificially constructed density-near-zero (DNZ) materials, we propose an alternative approach to realize a very robust acoustic cloaking, which includes one absorber and one projector connected by an energy channel. The elementary unit cell is made of one single-piece homogeneous copper, but it well maintains both the original planar wavefront and the nearly perfect one-dimensional transmission, in the presence of any inserted objects. Furthermore, the overall cloaked space can be increased, designed, and distributed by arbitrarily assembling additional DNZ cells without any upper limit in the total cloaking volume. We demonstrate acoustic cloaking by different combinations of DNZ cells in free space as well as in acoustic waveguides, which enables any objects inside the cells imperceptible along undistorted sound paths.

physics.class-ph↗

arXiv2013

Acoustic Wavefront Manipulation: Impedance Inhomogeneity and Extraordinary Reflection

Optical wavefront can be manipulated by interfering elementary beams with phase inhomogeneity. Therefore a surface allowing huge, abrupt and position-variant phase change would enable all possibilities of wavefront engineering. However, one may not have the luxury of efficient abrupt-phase-changing materials in acoustics. This motivates us to establish a counterpart mechanism for acoustics, in order to empower the wide spectrum of novel acoustic applications. Remarkably, the proposed impedance-governed generalized Snell's law (IGSL) of reflection is distinguished from that in optics. Via the manipulation of inhomogeneous acoustic impedance, extraordinary reflection can be tailored for unprecedented wavefront manipulation while ordinary reflection can be surprisingly switched on or off. Our results may power the acoustic-wave manipulation and engineering. We demonstrate novel acoustic applications by planar surfaces designed with IGSL.

physics.optics↗