arXiv subjects

Shijie Huang

Publications and source records attributed to Shijie Huang.

17 recordsLinked to original sources

arXiv2026

EDITBRIDGE: Towards Faithful and Efficient Ultra-High-Resolution Image Editing

High-resolution image editing is increasingly demanded in professional workflows, yet existing diffusion-based models remain constrained to resolutions below 1K due to quadratic attention complexity and prohibitive memory requirements. A prevalent workaround employs a two-stage pipeline: editing at low resolution followed by independent super-resolution. However, this approach suffers from two critical issues: information divergence, where hallucinated details contradict the original high-resolution (HR) source, and texture degradation, manifesting as over-smoothed or over-sharpened artifacts. We propose EditBridge, a diffusion bridge framework for efficient ultra high-resolution editing. Unlike conventional diffusion that regenerates from noise, we formulate refinement as structured data-to-data translation from the low-resolution (LR) edited result to its HR counterpart, explicitly conditioned on the original HR source to preserve authentic details. To efficiently incorporate HR source guidance, we introduce a prior-guided block-wise sparse attention mechanism that exploits semantic correspondence from first-stage editing to constrain cross-image interactions to spatially aligned regions, significantly reducing computational overhead. Extensive experiments demonstrate that EditBridge achieves high-fidelity editing with superior perceptual quality at resolutions up to 4K, delivering 3.6--8.4$\times$ speedup at 2K and enabling practical 4K editing in 61 seconds.

cs.CV↗

arXiv2026

Live Avatar: Streaming Real-time Audio-Driven Avatar Generation with Infinite Length

Audio-driven avatar interaction demands real-time, streaming, and infinite-length generation -- capabilities fundamentally at odds with the sequential denoising and long-horizon drift of current diffusion models. We present Live Avatar, an algorithm-system co-designed framework that addresses both challenges for a 14-billion-parameter diffusion model. On the algorithm side, a two-stage pipeline distills a pretrained bidirectional model into a causal, few-step streaming one, while a set of complementary long-horizon strategies eliminate identity drift and visual artifacts, enabling stable autoregressive generation exceeding 10000 seconds. On the system side, Timestep-forcing Pipeline Parallelism (TPP) assigns each GPU a fixed denoising timestep, converting the sequential diffusion chain into an asynchronous spatial pipeline that simultaneously boosts throughput and improves temporal consistency. Live Avatar achieves 45 FPS with a TTFF of 1.21\,s on 5 H800 GPUs, and to our knowledge is the first to enable practical real-time streaming of a 14B diffusion model for infinite-length avatar generation. We further introduce GenBench, a standardized long-form benchmark, to facilitate reproducible evaluation. Our project page is at https://liveavatar.github.io/.

cs.CV↗

arXiv2026

Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer

The landscape of high-performance image generation models is currently dominated by proprietary systems, such as Nano Banana Pro and Seedream 4.0. Leading open-source alternatives, including Qwen-Image, Hunyuan-Image-3.0 and FLUX.2, are characterized by massive parameter counts (20B to 80B), making them impractical for inference, and fine-tuning on consumer-grade hardware. To address this gap, we propose Z-Image, an efficient 6B-parameter foundation generative model built upon a Scalable Single-Stream Diffusion Transformer (S3-DiT) architecture that challenges the "scale-at-all-costs" paradigm. By systematically optimizing the entire model lifecycle -- from a curated data infrastructure to a streamlined training curriculum -- we complete the full training workflow in just 314K H800 GPU hours (approx. $630K). Our few-step distillation scheme with reward post-training further yields Z-Image-Turbo, offering both sub-second inference latency on an enterprise-grade H800 GPU and compatibility with consumer-grade hardware (<16GB VRAM). Additionally, our omni-pre-training paradigm also enables efficient training of Z-Image-Edit, an editing model with impressive instruction-following capabilities. Both qualitative and quantitative experiments demonstrate that our model achieves performance comparable to or surpassing that of leading competitors across various dimensions. Most notably, Z-Image exhibits exceptional capabilities in photorealistic image generation and bilingual text rendering, delivering results that rival top-tier commercial models, thereby demonstrating that state-of-the-art results are achievable with significantly reduced computational overhead. We publicly release our code, weights, and online demo to foster the development of accessible, budget-friendly, yet state-of-the-art generative models.

cs.CV↗

arXiv2026

Benchmarking Sequential Feedback Optimization for Wind Farm Power Maximization

This paper benchmarks sequential feedback optimization (SFO) for wind farm power maximization using a medium-fidelity dynamic flow model. We compare SFO with two well-established approaches, adjoint-based economic model predictive control (AMPC) and extremum seeking control (ESC), under a common nine-turbine layout and identical operating constraints. The comparison focuses on steady-state power production and computational efficiency, both relevant for real-time implementation. The simulation results illustrate that SFO achieves higher steady-state power while preserving real-time feasibility, AMPC provides a better transient performance at a higher online computational cost and without guarantees of convergence to the steady-state optimum, and ESC offers a computationally inexpensive model-free baseline that may converge to locally optimal solutions. These results provide a practical reference for selecting wind farm control strategies and for designing scalable, real-time optimization methods.

eess.SY↗

arXiv2026

CollectionLoRA: Collecting 50 Effects in 1 LoRA via Multi-Teacher On-Policy Distillation

Customized image editing aims to equip pre-trained diffusion models with specific visual effects using limited paired data, typically via Low-Rank Adaptation (LoRA). As the number of desired effects grows, storing and dynamically loading numerous these effect LoRAs significantly increases deployment overhead. Furthermore, current pipelines typically cascade these effect LoRAs with acceleration modules for fast generation, which triggers severe parameter interference and results in concept bleeding and style degradation. We propose CollectionLoRA, a multi-teacher on-policy distillation framework capable of distilling the concepts of up to 50 different effect LoRAs along with few-step generation capabilities into a single LoRA. This fundamentally resolves the feature interference issue and significantly reduces deployment costs. Specifically, the method introduces (i) a Probabilistic Dual-Stream Routing mechanism that enables the model to randomly switch between data sources during training, effectively enhancing its generalization in unseen scenarios; (ii) an Asymmetric Orthogonal Prompting strategy to achieve concept isolation within the prompt space; (iii) a Coarse-to-Fine Distillation Objective to mitigate the distribution gap between the teacher and student models. Extensive evaluations show that CollectionLoRA distills all customized effects and few-step generation into a single LoRA, reducing deployment overhead while achieving concept fidelity comparable to or better than independently trained teacher models. Code: https://github.com/Qwen-Applications/CollectionLoRA

cs.CV↗

arXiv2026

Open-Sora 2.0: Training a Commercial-Level Video Generation Model in $200k

Video generation models have achieved remarkable progress in the past year. The quality of AI video continues to improve, but at the cost of larger model size, increased data quantity, and greater demand for training compute. In this report, we present Open-Sora 2.0, a commercial-level video generation model trained for only $200k. With this model, we demonstrate that the cost of training a top-performing video generation model is highly controllable. We detail all techniques that contribute to this efficiency breakthrough, including data curation, model architecture, training strategy, and system optimization. According to human evaluation results and VBench scores, Open-Sora 2.0 is comparable to global leading video generation models including the open-source HunyuanVideo and the closed-source Runway Gen-3 Alpha. By making Open-Sora 2.0 fully open-source, we aim to democratize access to advanced video generation technology, fostering broader innovation and creativity in content creation. All resources are publicly available at: https://github.com/hpcaitech/Open-Sora.

cs.GR↗

arXiv2025

Benchmarking Retrieval-Augmented Multimodal Generation for Document Question Answering

Document Visual Question Answering (DocVQA) faces dual challenges in processing lengthy multimodal documents (text, images, tables) and performing cross-modal reasoning. Current document retrieval-augmented generation (DocRAG) methods remain limited by their text-centric approaches, frequently missing critical visual information. The field also lacks robust benchmarks for assessing multimodal evidence selection and integration. We introduce MMDocRAG, a comprehensive benchmark featuring 4,055 expert-annotated QA pairs with multi-page, cross-modal evidence chains. Our framework introduces innovative metrics for evaluating multimodal quote selection and enables answers that interleave text with relevant visual elements. Through large-scale experiments with 60 VLM/LLM models and 14 retrieval systems, we identify persistent challenges in multimodal evidence retrieval, selection, and integration.Key findings reveal advanced proprietary LVMs show superior performance than open-sourced alternatives. Also, they show moderate advantages using multimodal inputs over text-only inputs, while open-source alternatives show significant performance degradation. Notably, fine-tuned LLMs achieve substantial improvements when using detailed image descriptions. MMDocRAG establishes a rigorous testing ground and provides actionable insights for developing more robust multimodal DocVQA systems. Our benchmark and code are available at https://mmdocrag.github.io/MMDocRAG/.

cs.IR↗

arXiv2025

Sequential feedback optimization with application to wind farm control

This paper develops a sequential-linearization feedback optimization framework for driving nonlinear dynamical systems to an optimal steady state. A fundamental challenge in feedback optimization is the requirement of accurate first-order information of the steady-state input-output mapping, which is computationally prohibitive for high-dimensional nonlinear systems and often leads to poor performance when approximated around a fixed operating point. To address this limitation, we propose a sequential algorithm that adaptively updates the linearization point during optimization, maintaining local accuracy throughout the trajectory. We prove convergence to a neighborhood of the optimal steady state with explicit error bounds. To reduce the computational burden of repeated linearization operations, we further develop a multi-timescale variant where linearization updates occur at a slower timescale than optimization iterations, achieving significant computational savings while preserving convergence guarantees. The effectiveness of the proposed framework is demonstrated via numerical simulations of a realistic wind farm control problem. The results validate both the theoretical convergence predictions and the expected computational advantages of our multi-timescale formulation.

math.OC↗

arXiv2025

PhotoDoodle: Learning Artistic Image Editing from Few-Shot Pairwise Data

We introduce PhotoDoodle, a novel image editing framework designed to facilitate photo doodling by enabling artists to overlay decorative elements onto photographs. Photo doodling is challenging because the inserted elements must appear seamlessly integrated with the background, requiring realistic blending, perspective alignment, and contextual coherence. Additionally, the background must be preserved without distortion, and the artist's unique style must be captured efficiently from limited training data. These requirements are not addressed by previous methods that primarily focus on global style transfer or regional inpainting. The proposed method, PhotoDoodle, employs a two-stage training strategy. Initially, we train a general-purpose image editing model, OmniEditor, using large-scale data. Subsequently, we fine-tune this model with EditLoRA using a small, artist-curated dataset of before-and-after image pairs to capture distinct editing styles and techniques. To enhance consistency in the generated results, we introduce a positional encoding reuse mechanism. Additionally, we release a PhotoDoodle dataset featuring six high-quality styles. Extensive experiments demonstrate the advanced performance and robustness of our method in customized image editing, opening new possibilities for artistic creation.

cs.CV↗

arXiv2024

DVasMesh: Deep Structured Mesh Reconstruction from Vascular Images for Dynamics Modeling of Vessels

Vessel dynamics simulation is vital in studying the relationship between geometry and vascular disease progression. Reliable dynamics simulation relies on high-quality vascular meshes. Most of the existing mesh generation methods highly depend on manual annotation, which is time-consuming and laborious, usually facing challenges such as branch merging and vessel disconnection. This will hinder vessel dynamics simulation, especially for the population study. To address this issue, we propose a deep learning-based method, dubbed as DVasMesh to directly generate structured hexahedral vascular meshes from vascular images. Our contributions are threefold. First, we propose to formally formulate each vertex of the vascular graph by a four-element vector, including coordinates of the centerline point and the radius. Second, a vectorized graph template is employed to guide DVasMesh to estimate the vascular graph. Specifically, we introduce a sampling operator, which samples the extracted features of the vascular image (by a segmentation network) according to the vertices in the template graph. Third, we employ a graph convolution network (GCN) and take the sampled features as nodes to estimate the deformation between vertices of the template graph and target graph, and the deformed graph template is used to build the mesh. Taking advantage of end-to-end learning and discarding direct dependency on annotated labels, our DVasMesh demonstrates outstanding performance in generating structured vascular meshes on cardiac and cerebral vascular images. It shows great potential for clinical applications by reducing mesh generation time from 2 hours (manual) to 30 seconds (automatic).

eess.IV↗

arXiv2024

HiFAST: An HI Data Calibration and Imaging Pipeline for FAST III. Standing Wave Removal

The standing waves existed in radio telescope data are primarily due to reflections among the instruments, which significantly impact the spectrum quality of the Five-hundred-meter Aperture Spherical radio Telescope (FAST). Eliminating these standing waves for FAST is challenging given the constant changes in their phases and amplitudes. Over a ten-second period, the phases shift by 18$^{\circ}$ while the amplitudes fluctuate by 6 mK. Thus, we developed the fast Fourier transform (FFT) filter method to eliminate these standing waves for every individual spectrum. The FFT filter can decrease the root mean square (RMS) from 3.2 to 1.15 times the theoretical estimate. Compared to other methods such as sine fitting and running median, the FFT filter achieves a median RMS of approximately 1.2 times the theoretical expectation and the smallest scatter at 12%. Additionally, the FFT filter method avoids the flux loss issue encountered with some other methods. The FFT is also efficient in detecting harmonic radio frequency interference (RFI). In the FAST data, we identified three distinct types of harmonic RFI, each with amplitudes exceeding 100 mK and intrinsic frequency periods of 8.1, 0.5, and 0.37 MHz, respectively. The FFT filter, proven as the most effective method, is integrated into the HI data calibration and imaging pipeline for FAST (HiFAST, https://hifast.readthedocs.io).

astro-ph.IM↗

arXiv2024

ProcessPainter: Learn Painting Process from Sequence Data

The painting process of artists is inherently stepwise and varies significantly among different painters and styles. Generating detailed, step-by-step painting processes is essential for art education and research, yet remains largely underexplored. Traditional stroke-based rendering methods break down images into sequences of brushstrokes, yet they fall short of replicating the authentic processes of artists, with limitations confined to basic brushstroke modifications. Text-to-image models utilizing diffusion processes generate images through iterative denoising, also diverge substantially from artists' painting process. To address these challenges, we introduce ProcessPainter, a text-to-video model that is initially pre-trained on synthetic data and subsequently fine-tuned with a select set of artists' painting sequences using the LoRA model. This approach successfully generates painting processes from text prompts for the first time. Furthermore, we introduce an Artwork Replication Network capable of accepting arbitrary-frame input, which facilitates the controlled generation of painting processes, decomposing images into painting sequences, and completing semi-finished artworks. This paper offers new perspectives and tools for advancing art education and image generation technology.

cs.CV↗

arXiv2023

Tissue Segmentation of Thick-Slice Fetal Brain MR Scans with Guidance from High-Quality Isotropic Volumes

Accurate tissue segmentation of thick-slice fetal brain magnetic resonance (MR) scans is crucial for both reconstruction of isotropic brain MR volumes and the quantification of fetal brain development. However, this task is challenging due to the use of thick-slice scans in clinically-acquired fetal brain data. To address this issue, we propose to leverage high-quality isotropic fetal brain MR volumes (and also their corresponding annotations) as guidance for segmentation of thick-slice scans. Due to existence of significant domain gap between high-quality isotropic volume (i.e., source data) and thick-slice scans (i.e., target data), we employ a domain adaptation technique to achieve the associated knowledge transfer (from high-quality volumes to thick-slice scans). Specifically, we first register the available high-quality isotropic fetal brain MR volumes across different gestational weeks to construct longitudinally-complete source data. To capture domain-invariant information, we then perform Fourier decomposition to extract image content and style codes. Finally, we propose a novel Cycle-Consistent Domain Adaptation Network (C2DA-Net) to efficiently transfer the knowledge learned from high-quality isotropic volumes for accurate tissue segmentation of thick-slice scans. Our C2DA-Net can fully utilize a small set of annotated isotropic volumes to guide tissue segmentation on unannotated thick-slice scans. Extensive experiments on a large-scale dataset of 372 clinically acquired thick-slice MR scans demonstrate that our C2DA-Net achieves much better performance than cutting-edge methods quantitatively and qualitatively.

eess.IV↗

arXiv2023

Categorize Radio Interference using component and temporal analysis

Radio frequency interference (RFI) is a significant challenge faced by today's radio astronomers. While most past efforts were devoted to cleaning the RFI from the data, we develop a novel method for categorizing and cataloguing RFI for forensic purpose. We present a classifier that categorizes RFI into different types based on features extracted using Principal Component Analysis (PCA) and Fourier analysis. The classifier can identify narrowband non-periodic RFI above 2 sigma, narrowband periodic RFI above 3 sigma, and wideband impulsive RFI above 5 sigma with F1 scores between 0.87 and 0.91 in simulation. This classifier could be used to identify the sources of RFI as well as to clean RFI contamination (particularly in pulsar search). In the long-term analysis of the categorized RFI, we found a special type of drifting periodic RFI that is detrimental to pulsar search. We also found evidences of an increased rate of impulsive RFI when the telescope is pointing toward the cities. These results demonstrate this classifier's potential as a forensic tool for RFI environment monitoring of radio telescopes.

astro-ph.IM↗

arXiv2023

Distributed Non-Bayesian Learning for Games with Incomplete Information

We consider distributed learning problem in games with an unknown cost-relevant parameter, and aim to find the Nash equilibrium while learning the true parameter. Inspired by the social learning literature, we propose a distributed non-Bayesian rule to learn the parameter (each agent maintains a belief on the parameter and updates the belief according to the received noisy cost), combined with best response dynamics for strategy update. The difficulty of the analysis lies mainly in the fact that the parameter learning process and strategy update process are coupled. We first prove that agents' beliefs converge to a common belief and the strategy profiles converge to a Nash equilibrium under this common belief. On this basis, we further show that the beliefs eventually concentrate on the true parameter.

math.OC↗

arXiv2022

No-Regret Learning in Network Stochastic Zero-Sum Games

No-regret learning has been widely used to compute a Nash equilibrium in two-person zero-sum games. However, there is still a lack of regret analysis for network stochastic zero-sum games, where players competing in two subnetworks only have access to some local information, and the cost functions include uncertainty. Such a game model can be found in security games, when a group of inspectors work together to detect a group of evaders. In this paper, we propose a distributed stochastic mirror descent (D-SMD) method, and establish the regret bounds $O(\sqrt{T})$ and $O(\log T)$ in the expected sense for convex-concave and strongly convex-strongly concave costs, respectively. Our bounds match those of the best known first-order online optimization algorithms. We then prove the convergence of the time-averaged iterates of D-SMD to the set of Nash equilibria. Finally, we show that the actual iterates of D-SMD almost surely converge to the Nash equilibrium in the strictly convex-strictly concave setting.

math.OC↗

arXiv2021

No-regret distributed learning in subnetwork zero-sum games

In this paper, we consider a distributed learning problem in a subnetwork zero-sum game, where agents are competing in different subnetworks. These agents are connected through time-varying graphs where each agent has its own cost function and can receive information from its neighbors. We propose a distributed mirror descent algorithm for computing a Nash equilibrium and establish a sublinear regret bound on the sequence of iterates when the graphs are uniformly strongly connected and the cost functions are convex-concave. Moreover, we prove its convergence with suitably selected diminishing stepsizes for a strictly convex-concave cost function. We also consider a constant step-size variant of the algorithm and establish an asymptotic error bound between the cost function values of running average actions and a Nash equilibrium. In addition, we apply the algorithm to compute a mixed-strategy Nash equilibrium in subnetwork zero-sum finite-strategy games, which have merely convex-concave (to be specific, multilinear) cost functions, and obtain a final-iteration convergence result and an ergodic convergence result, respectively, under different assumptions.

math.OC↗