arXiv subjects

Hao Meng

Publications and source records attributed to Hao Meng.

At least 19 recordsLinked to original sources

arXiv2026

Qwen-Image-2.0-RL Technical Report

We present Qwen-Image-2.0-RL, a post-training pipeline that applies reinforcement learning from human feedback (RLHF) and on-policy distillation (OPD) to improve both the visual quality and instruction-following capability of the Qwen-Image-2.0 diffusion model. To provide reliable reward signals, we construct task-specific composite reward models by fine-tuning vision-language models with a pointwise scoring paradigm and chain-of-thought reasoning. For text-to-image generation, the reward models cover alignment, aesthetics, and portrait fidelity dimensions. For image editing tasks, the reward system addresses instruction-following accuracy and face identity preservation. Building on this reward system, we develop a scalable GRPO-based RL training framework, incorporating a hybrid classifier-free guidance (CFG) strategy to preserve pre-trained knowledge, prompt curation via intra-group reward range filtering, and per-category reward weight calibration. To merge the task-specialized RL policies for T2I and editing, we propose on-policy distillation as the final training stage, which consolidates multiple teachers into a single student model through trajectory-level velocity matching. Extensive evaluation shows that Qwen-Image-2.0-RL achieves 57.84 overall score on Qwen-Image-Bench (+2.61 over the base model), Elo ratings of 1193 in text-to-image arena (+78) and 1349 in image edit arena (+93), demonstrating consistent gains in aesthetic quality, prompt adherence, and editing accuracy.

cs.CV↗

arXiv2026

SoulX-Transcriber: A Robust End-to-End Framework for Multi-Speaker Speech Transcription

Recent advances in Automatic Speech Recognition (ASR) and Large Language Models (LLMs) have significantly improved speech understanding capabilities. However, multi-speaker speech transcription remains challenging task, constrained by highly similar speaker voices, rapid turn-taking transitions, overlapping utterances and inaccurate speaker boundary segmentation. These challenges become particularly pronounced in real-world conversational audio, where speaker dynamics and acoustic conditions are highly variable. This technical report presents SoulX-Transcriber, a unified multi-speaker transcription system that jointly models speaker diarization (SD) and ASR within an LLM-based framework. SoulX-Transcriber adopts a two-stage training strategy to improve both speaker discrimination and transcription robustness. In the first stage, speaker-aware multi-task continuous pre-training enhances speaker representation learning and boundary perception. In the second stage, supervised fine-tuning further optimizes the model for accurate end-to-end speaker-attributed transcription under complex multi-speaker conditions. SoulX-Transcriber delivers strong performance and robustness across multiple public benchmarks, including AliMeeting, AISHELL-4, and AMI, while maintaining high adaptability to multi-domain scenarios.

eess.AS↗

arXiv2026

Qwen-Image-VAE-2.0 Technical Report

We present Qwen-Image-VAE-2.0, a suite of high-compression Variational Autoencoders (VAEs) that achieve significant advances in both reconstruction fidelity and diffusability. To address the reconstruction bottlenecks of high compression, we adopt an improved architecture featuring Global Skip Connections (GSC) and expanded latent channels. Moreover, we scale training to billions of images and incorporate a synthetic rendering engine to improve performance in text-rich scenarios. To tackle the convergence challenges of high-dimensional latent space, we implement an enhanced semantic alignment strategy to make the latent space highly amenable to diffusion modeling. To optimize computational efficiency, we leverage an asymmetric and attention-free encoder-decoder backbone to minimize encoding overhead. We present a comprehensive evaluation of Qwen-Image-VAE-2.0 on public reconstruction benchmarks. To evaluate performance in text-rich scenarios, we propose OmniDoc-TokenBench, a new benchmark comprising a diverse collection of real-world documents coupled with specialized OCR-based evaluation metrics. Qwen-Image-VAE-2.0 achieves state-of-the-art reconstruction performance, demonstrating exceptional capabilities in both general domains and text-rich scenarios at high compression ratio. Furthermore, downstream DiT experiments reveal our models possess superior diffusability, significantly accelerating convergence compared to existing high-compression baselines. These establish Qwen-Image-VAE-2.0 as a leading model with high compression, superior reconstruction, and exceptional diffusability.

cs.CV↗

arXiv2026

Qwen-Image-2.0 Technical Report

We present Qwen-Image-2.0, an omni-capable image generation foundation model that unifies high-fidelity generation and precise image editing within a single framework. Despite recent progress, existing models still struggle with ultra-long text rendering, multilingual typography, high-resolution photorealism, robust instruction following, and efficient deployment, especially in text-rich and compositionally complex scenarios. Qwen-Image-2.0 addresses these challenges by coupling Qwen3-VL as the condition encoder with a Multimodal Diffusion Transformer for joint condition-target modeling, supported by large-scale data curation and a customized multi-stage training pipeline. This enables strong multimodal understanding while preserving flexible generation and editing capabilities. The model supports instructions of up to 1K tokens for generating text-rich content such as slides, posters, infographics, and comics, while significantly improving multilingual text fidelity and typography. It also enhances photorealistic generation with richer details, more realistic textures, and coherent lighting, and follows complex prompts more reliably across diverse styles. Extensive human evaluations show that Qwen-Image-2.0 substantially outperforms previous Qwen-Image models in both generation and editing, marking a step toward more general, reliable, and practical image generation foundation models.

cs.CV↗

arXiv2026

Aligning Language Models for Lyric-to-Melody Generation with Rule-Based Musical Constraints

Large Language Models (LLMs) show promise in lyric-to-melody generation, but models trained with Supervised Fine-Tuning (SFT) often produce musically implausible melodies with issues like poor rhythm and unsuitable vocal ranges, a phenomenon we term "constraint violation". To address this, we propose a novel alignment framework that instills musical knowledge without human annotation. We define rule-based musical constraints to automatically generate a preference dataset from an SFT model's outputs. The model is then aligned through a sequential process, first using Direct Preference Optimization (DPO) on paired preference data, followed by Kahneman-Tversky Optimization (KTO) on unpaired negative samples. Experimental results demonstrate that our aligned model substantially reduces rule violations and outperforms strong baselines in both objective and subjective evaluations, generating melodies with substantially improved musicality and coherence. An interactive demo with audio comparisons is available at https://arain233.github.io/AligningMelody-demo.

cs.SD↗

arXiv2026

TIGFlow-GRPO: Trajectory Forecasting via Interaction-Aware Flow Matching and Reward-Guided Optimization

Human trajectory forecasting is important for intelligent multimedia systems operating in visually complex environments, such as autonomous driving and crowd surveillance. Although Conditional Flow Matching (CFM) has shown strong ability in modeling trajectory distributions from spatio-temporal observations, existing approaches still focus primarily on supervised fitting, which may leave social norms and scene constraints insufficiently reflected in generated trajectories. To address this issue, we propose TIGFlow-GRPO, a two-stage generative approach that aligns flow-based trajectory generation with behavioral rules. In the first stage, we build a CFM-based predictor with a Trajectory-Interaction-Graph (TIG) module to model fine-grained visual-spatial interactions and strengthen context encoding. This stage captures both agent-agent and agent-scene relations more effectively, providing more informative conditional features for subsequent alignment. In the second stage, we perform Flow-GRPO post-training, where deterministic flow rollout is reformulated as stochastic ODE-to-SDE sampling to enable trajectory exploration, and a composite reward combines view-aware social compliance with map-aware physical feasibility. By evaluating trajectories explored through SDE rollout, GRPO progressively steers multimodal predictions toward behaviorally plausible futures. Experiments on the ETH/UCY and SDD datasets show that TIGFlow-GRPOimproves forecasting accuracy and long-horizon stability while generatingtrajectories that are more socially compliant and physically feasible.These results suggest that the proposed approach provides an effective way to connectflow-based trajectory modeling with behavior-aware alignment in dynamic multimedia environments.

cs.CV↗

arXiv2026

Joint Learning Global-Local Speaker Classification to Enhance End-to-End Speaker Diarization and Recognition

Large Audio-Language Models (LALMs) have demonstrated remarkable performance in end-to-end speaker diarization and recognition. However, their speaker discriminability remains limited due to the scarcity of large-scale conversational data and the absence of explicit speaker representation optimization. To address this, we propose GLSC-SDR, a paradigm that jointly trains speaker classification with diarization and recognition. We further introduce a Global-Local Speaker Classification strategy, which uses clustered speakers as global labels and re-encoded intra-cluster speakers as local labels. This hierarchical design enhances fine-grained speaker discrimination while preserving semantic transcription accuracy. Experiments on AliMeeting, AISHELL-4, and AMI-SDM demonstrate that GLSC-SDR achieves competitive or superior performance compared to simulation-based and multi-encoder approaches, without relying on large-scale real conversational data.

cs.SD↗

arXiv2026

SoulX-Duplug: Plug-and-Play Streaming State Prediction Module for Realtime Full-Duplex Speech Conversation

Recent advances in spoken dialogue systems have brought increased attention to human-like full-duplex voice interactions. However, our comprehensive review of this field reveals several challenges, including the difficulty in obtaining training data, catastrophic forgetting, and limited scalability. In this work, we propose SoulX-Duplug, a plug-and-play streaming state prediction module for full-duplex spoken dialogue systems. By jointly performing streaming ASR, SoulX-Duplug explicitly leverages textual information to identify user intent, effectively serving as a semantic VAD. To promote fair evaluation, we introduce SoulX-Duplug-Eval, extending widely used benchmarks with improved bilingual coverage. Experimental results show that SoulX-Duplug enables low-latency streaming dialogue state control, and the system built upon it outperforms existing full-duplex models in overall turn management and latency performance. We have open-sourced SoulX-Duplug and SoulX-Duplug-Eval.

eess.AS↗

arXiv2026

Inverse orbital Hall effect induced terahertz emission enabled by a ferromagnet with quenched orbital moment in Fe/Pt/W trilayers

The inverse orbital Hall effect (IOHE) has recently attracted considerable attention as an emerging mechanism for terahertz (THz) emission based on ultrafast angular-momentum-to-charge conversion. Most experimental studies have focused on materials with strong spin-orbit coupling or pronounced orbital character, where sizable orbital Hall responses are expected. Elemental ferromagnets such as Fe are generally regarded as quenched orbital sources and are not expected to exhibit orbital-dominated THz emission. Here, we report a pronounced enhancement of THz emission in Fe/Pt/W trilayer heterostructures, despite the absence of detectable orbital contributions in the corresponding Fe/Pt and Fe/W bilayers. Thickness-dependent measurements reveal long-distance signal persistence, systematic delay accumulation, and pronounced pulse broadening with increasing W thickness. These features are inconsistent with diffusive spin transport and indicate that orbital angular momentum transport in the W layer, converted into charge current via the IOHE, becomes a dominant channel for THz emission in the trilayer configuration. Our results demonstrate that strong IOHE can emerge in heterostructures incorporating a quenched orbital ferromagnet, providing an effective route to enhance spintronic THz emitters through orbital Hall physics.

cond-mat.mtrl-sci↗

arXiv2026

SoulX-Singer: Towards High-Quality Zero-Shot Singing Voice Synthesis

While recent years have witnessed rapid progress in speech synthesis, open-source singing voice synthesis (SVS) systems still face significant barriers to industrial deployment, particularly in terms of robustness and zero-shot generalization. In this report, we introduce SoulX-Singer, a high-quality open-source SVS system designed with practical deployment considerations in mind. SoulX-Singer supports controllable singing generation conditioned on either symbolic musical scores (MIDI) or melodic representations, enabling flexible and expressive control in real-world production workflows. Trained on more than 42,000 hours of vocal data, the system supports Mandarin Chinese, English, and Cantonese and consistently achieves state-of-the-art synthesis quality across languages under diverse musical conditions. Furthermore, to enable reliable evaluation of zero-shot SVS performance in practical scenarios, we construct SoulX-Singer-Eval, a dedicated benchmark with strict training-test disentanglement, facilitating systematic assessment in zero-shot settings.

eess.AS↗

arXiv2026

Probing Dark Matter Halos of High-redshift Quasars via Wide-Field Clustering

High-redshift quasars are powerful tracers of both astrophysical processes and large-scale structure in the early Universe. Using a sample of 1,251 spectroscopically confirmed quasars and 827 highly reliable photometric quasar candidates selected with a machine-learning framework, all at $4.7 \leq z < 6.3$ (excluding $5.4 < z < 5.6$) and with a median luminosity of $M_{1450}\sim -25.5$, we investigate the large-scale environments of quasars near the end of cosmic reionization. We measure the projected auto-correlation function of the quasar population and derive bias parameters of $b=22.11^{+2.17}_{-2.19}$ and $28.55^{+5.76}_{-5.75}$ for the redshift intervals $4.7\leq z<5.4$ and $5.6\leq z<6.3$, respectively. These correspond to characteristic dark matter halo masses of $\log(M_{ h}/M_\odot)=12.66^{+0.10}_{-0.11}$ and $12.62^{+0.21}_{-0.21}$, respectively. We further estimate quasar duty cycles of $0.011^{+0.017}_{-0.007}$ and $0.012^{+0.104}_{-0.011}$ for the two redshift bins. These measurements are consistent with previous studies and the observed $f_{\rm duty}$--$M_{\rm halo}$ relation, indicating that only a small fraction of suitable dark matter halos host optically luminous quasars at any given time. The inferred low duty cycles suggest that a substantial fraction of supermassive black hole growth may occur during obscured accretion phases. Our results provide new constraints on the connection between high-redshift quasars and their host dark matter halos, offering valuable insights into the co-evolution of supermassive black holes and large-scale structure in the early Universe.

astro-ph.GA↗

arXiv2025

Qwen-Image Technical Report

We present Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing. To address the challenges of complex text rendering, we design a comprehensive data pipeline that includes large-scale data collection, filtering, annotation, synthesis, and balancing. Moreover, we adopt a progressive training strategy that starts with non-text-to-text rendering, evolves from simple to complex textual inputs, and gradually scales up to paragraph-level descriptions. This curriculum learning approach substantially enhances the model's native text rendering capabilities. As a result, Qwen-Image not only performs exceptionally well in alphabetic languages such as English, but also achieves remarkable progress on more challenging logographic languages like Chinese. To enhance image editing consistency, we introduce an improved multi-task training paradigm that incorporates not only traditional text-to-image (T2I) and text-image-to-image (TI2I) tasks but also image-to-image (I2I) reconstruction, effectively aligning the latent representations between Qwen2.5-VL and MMDiT. Furthermore, we separately feed the original image into Qwen2.5-VL and the VAE encoder to obtain semantic and reconstructive representations, respectively. This dual-encoding mechanism enables the editing module to strike a balance between preserving semantic consistency and maintaining visual fidelity. Qwen-Image achieves state-of-the-art performance, demonstrating its strong capabilities in both image generation and editing across multiple benchmarks.

cs.CV↗

arXiv2025

Quantized resonant tunneling effect in Josephson junctions with ferromagnetic bilayers

We study the Josephson effect in one-dimensional SF$_1$F$_2$S junctions, which consist of conventional s-wave superconductors (S) connected by two ferromagnetic layers (F$_1$ and F$_2$). At low temperatures, the potential barrier at the F$_1$/F$_2$ interface can induce a quantized resonant tunneling effect. This effect not only modifies the amplitude of the critical current but also affects the phase of the Josephson current. As the exchange fields ($h_1$, $h_2$) and thicknesses ($d_1$, $d_2$) of the F$_1$ and F$_2$ layers vary, the critical current displays periodic resonance peaks. These peaks occur under the quantization conditions $Q_{1(2)} d_{1(2)} = \left(n_{1(2)} + 1/2\right) \pi$, where $Q_{1(2)} = 2h_{1(2)}/(\hbar v_F)$ is the center-of-mass momentum carried by Cooper pairs, with $v_F$ being the Fermi velocity, and $n_{1(2)} = 0, 1, 2, \cdots$. It can be inferred that the potential barrier suppresses the transport of spin-singlet pairs while allowing spin-triplet pairs with zero spin projection along the magnetization axis to pass through. As these spin-triplet pairs traverse the F$_1$ and F$_2$ layers, the total phase they acquire determines the ground state of the Josephson junction. At the resonance peaks, the Josephson current primarily arises from the first harmonic in both the parallel and antiparallel magnetization configurations. However, in perpendicular configurations, the second harmonic becomes more significant. In scenarios where both ferromagnetic layers have identical exchange fields and thicknesses, the potential barrier selectively suppresses the current in the 0-state while allowing it to persist in the $\pi$-state for parallel configurations. Conversely, in antiparallel configurations, the current in the 0-state is consistently preserved.

cond-mat.supr-con↗

arXiv2025

Video Deblurring by Sharpness Prior Detection and Edge Information

Video deblurring is essential task for autonomous driving, facial recognition, and security surveillance. Traditional methods directly estimate motion blur kernels, often introducing artifacts and leading to poor results. Recent approaches utilize the detection of sharp frames within video sequences to enhance deblurring. However, existing datasets rely on fixed number of sharp frames, which may be too restrictive for some applications and may introduce a bias during model training. To address these limitations and enhance domain adaptability, this work first introduces GoPro Random Sharp (GoProRS), a new dataset where the the frequency of sharp frames within the sequence is customizable, allowing more diverse training and testing scenarios. Furthermore, it presents a novel video deblurring model, called SPEINet, that integrates sharp frame features into blurry frame reconstruction through an attention-based encoder-decoder architecture, a lightweight yet robust sharp frame detection and an edge extraction phase. Extensive experimental results demonstrate that SPEINet outperforms state-of-the-art methods across multiple datasets, achieving an average of +3.2% PSNR improvement over recent techniques. Given such promising results, we believe that both the proposed model and dataset pave the way for future advancements in video deblurring based on the detection of sharp frames.

cs.CV↗

arXiv2022

Anomalous supercurrent modulated by interfacial magnetizations in Josephson junctions with ferromagnetic bilayers

Based on the Bogoliubov-de Gennes equations, we investigate the transport of the Josephson current in a S/$f_L$-F$_1$-$f_C$-F$_2$-$f_R$/S junction, where S and F$_{1,2}$ are superconductors and ferromagnets, and $f_{L, C, R}$ are the left, central, and right spin-active interfaces. These interfaces have noncollinear magnetizations, and the azimuthal angles of the magnetizations at the $f_{L, C, R}$ interfaces are $χ_{L, C, R}$. We demonstrate that, if both the ferromagnets have antiparallel magnetizations, the critical current oscillates as a function of the exchange field and the thickness of the ferromagnets for particular $χ_L$ or $χ_R$. By contrast, when the magnetization at the $f_C$ interface is perpendicular to that at the $f_L$ and $f_R$ interfaces, the critical current reaches a larger value and is hardly affected by the exchange field and the thickness. Interestingly, if both the ferromagnets are converted to antiparallel half-metals, the critical current maintains a constant value and rarely changes with the ferromagnetic thicknesses and the azimuthal angles. At this time, an anomalous supercurrent can appear in the system, in which case the Josephson current still exists even if the superconducting phase difference $ϕ$ is zero. This supercurrent satisfies the current-phase relation $I=I_c\sin(ϕ+ϕ_0)$ with $I_c$ being the critical current and $ϕ_0=2χ_C-χ_L-χ_R$. We deduce that the additional phase $ϕ_0$ arises from phase superposition, where the phase is captured by the spin-triplet pairs when they pass through each spin-active interface. In addition, when both the ferromagnets are transformed into parallel half-metals, the $f_C$ interface never contributes any phase to the supercurrent and $ϕ_0=χ_R-χ_L+π$. In such a case, the current-phase relation is similar to that in a S/$f_L$-F-$f_R$/S junction.

cond-mat.supr-con↗

arXiv2022

Current-driven magnetization dynamics and their correlation with magnetization configurations in perpendicularly magnetized tunnel junctions

We study spin-transfer-torque driven magnetization dynamics of a perpendicular magnetic tunnel junction (MTJ) nanopillar. Based on the combination of spin-torque ferromagnetic resonance and microwave spectroscopy techniques, we demonstrate that the free layer (FL) and the weak pinned reference layer (RL) exhibit distinct dynamic behaviors with opposite frequency vs. field dispersion relations. The FL can support a single coherent spin-wave (SW) mode for both parallel and antiparallel configurations, while the RL exhibits spin-wave excitation only for the antiparallel state. These two SW modes corresponding to the FL and RL coexist at an antiparallel state and exhibit a crossover phenomenon of oscillation frequency with increasing the external magnetic field, which could be helpful in the mutual synchronization of auto-oscillations for SW-based neuromorphic computing.

cond-mat.mes-hall↗

arXiv2022

3D Interacting Hand Pose Estimation by Hand De-occlusion and Removal

Estimating 3D interacting hand pose from a single RGB image is essential for understanding human actions. Unlike most previous works that directly predict the 3D poses of two interacting hands simultaneously, we propose to decompose the challenging interacting hand pose estimation task and estimate the pose of each hand separately. In this way, it is straightforward to take advantage of the latest research progress on the single-hand pose estimation system. However, hand pose estimation in interacting scenarios is very challenging, due to (1) severe hand-hand occlusion and (2) ambiguity caused by the homogeneous appearance of hands. To tackle these two challenges, we propose a novel Hand De-occlusion and Removal (HDR) framework to perform hand de-occlusion and distractor removal. We also propose the first large-scale synthetic amodal hand dataset, termed Amodal InterHand Dataset (AIH), to facilitate model training and promote the development of the related research. Experiments show that the proposed method significantly outperforms previous state-of-the-art interacting hand pose estimation approaches. Codes and data are available at https://github.com/MengHao666/HDR.

cs.CV↗

arXiv2020

Spin-wave focusing induced skyrmion generation

We propose a new method to generate magnetic skyrmions through spin-wave focusing in chiral ferromagnets.A lens is constructed to focus spin waves by a curved interface between two ferromagnetic thin films with different perpendicular magnetic anisotropies. Based on the principle of identical magnonic path length, we derive the lens contour that can be either elliptical or hyperbolical depending on the magnon refractive index. Micromagnetic simulations are performed to verify the theoretical design. It is found that under proper condition magnetic skyrmions emerge near the focus point of the lens where the spin-wave intensity has been significantly enhanced. A close investigation shows that a magnetic droplet first forms and then converts to the skyrmion accompanying with a change of topological charge. Phase diagram about the amplitude and duration-time of the exciting field for skyrmion generation is obtained. Our findings would be helpful for designing novel spintronic devices combining the advantages of skyrmionics and magnonics.

cond-mat.mes-hall↗