Yi Yuan

Xubo Liu, Qiuqiang Kong, Yan Zhao et al. · bytedance

Language-queried audio source separation (LASS) is a new paradigm for computational auditory scene analysis (CASA). LASS aims to separate a target sound from an audio mixture given a natural language query, which provides a natural and scalable interface for digital audio applications. Recent works on LASS, despite attaining promising separation performance on specific sources (e.g., musical instruments, limited classes of audio events), are unable to separate audio concepts in the open domain. In this work, we introduce AudioSep, a foundation model for open-domain audio source separation with natural language queries. We train AudioSep on large-scale multimodal datasets and extensively evaluate its capabilities on numerous tasks including audio event separation, musical instrument separation, and speech enhancement. AudioSep demonstrates strong separation performance and impressive zero-shot generalization ability using audio captions or text labels as queries, substantially outperforming previous audio-queried and language-queried sound separation models. For reproducibility of this work, we will release the source code, evaluation benchmark and pre-trained model at: https://github.com/Audio-AGI/AudioSep.

Haohe Liu, Zehua Chen, Yi Yuan et al.

Text-to-audio (TTA) system has recently gained attention for its ability to synthesize general audio based on text descriptions. However, previous studies in TTA have limited generation quality with high computational costs. In this study, we propose AudioLDM, a TTA system that is built on a latent space to learn the continuous audio representations from contrastive language-audio pretraining (CLAP) latents. The pretrained CLAP models enable us to train LDMs with audio embedding while providing text embedding as a condition during sampling. By learning the latent representations of audio signals and their compositions without modeling the cross-modal relationship, AudioLDM is advantageous in both generation quality and computational efficiency. Trained on AudioCaps with a single GPU, AudioLDM achieves state-of-the-art TTA performance measured by both objective and subjective metrics (e.g., frechet distance). Moreover, AudioLDM is the first TTA system that enables various text-guided audio manipulations (e.g., style transfer) in a zero-shot fashion. Our implementation and demos are available at https://audioldm.github.io.

Haohe Liu, Yi Yuan, Xubo Liu et al.

Although audio generation shares commonalities across different types of audio, such as speech, music, and sound effects, designing models for each type requires careful consideration of specific objectives and biases that can significantly differ from those of other types. To bring us closer to a unified perspective of audio generation, this paper proposes a framework that utilizes the same learning method for speech, music, and sound effect generation. Our framework introduces a general representation of audio, called "language of audio" (LOA). Any audio can be translated into LOA based on AudioMAE, a self-supervised pre-trained representation learning model. In the generation process, we translate any modalities into LOA by using a GPT-2 model, and we perform self-supervised audio generation learning with a latent diffusion model conditioned on LOA. The proposed framework naturally brings advantages such as in-context learning abilities and reusable self-supervised pretrained AudioMAE and latent diffusion models. Experiments on the major benchmarks of text-to-audio, text-to-music, and text-to-speech demonstrate state-of-the-art or competitive performance against previous approaches. Our code, pretrained model, and demo are available at https://audioldm.github.io/audioldm2.

Yi Yuan, Haohe Liu, Xubo Liu et al.

Despite recent progress in text-to-audio (TTA) generation, we show that the state-of-the-art models, such as AudioLDM, trained on datasets with an imbalanced class distribution, such as AudioCaps, are biased in their generation performance. Specifically, they excel in generating common audio classes while underperforming in the rare ones, thus degrading the overall generation performance. We refer to this problem as long-tailed text-to-audio generation. To address this issue, we propose a simple retrieval-augmented approach for TTA models. Specifically, given an input text prompt, we first leverage a Contrastive Language Audio Pretraining (CLAP) model to retrieve relevant text-audio pairs. The features of the retrieved audio-text data are then used as additional conditions to guide the learning of TTA models. We enhance AudioLDM with our proposed approach and denote the resulting augmented system as Re-AudioLDM. On the AudioCaps dataset, Re-AudioLDM achieves a state-of-the-art Frechet Audio Distance (FAD) of 1.37, outperforming the existing approaches by a large margin. Furthermore, we show that Re-AudioLDM can generate realistic audio for complex scenes, rare audio classes, and even unseen audio types, indicating its potential in TTA tasks.

Xubo Liu, Zhongkai Zhu, Haohe Liu et al.

Despite breakthroughs in audio generation models, their capabilities are often confined to domain-specific conditions such as speech transcriptions and audio captions. However, real-world audio creation aims to generate harmonious audio containing various elements such as speech, music, and sound effects with controllable conditions, which is challenging to address using existing audio generation systems. We present WavJourney, a novel framework that leverages Large Language Models (LLMs) to connect various audio models for audio creation. WavJourney allows users to create storytelling audio content with diverse audio elements simply from textual descriptions. Specifically, given a text instruction, WavJourney first prompts LLMs to generate an audio script that serves as a structured semantic representation of audio elements. The audio script is then converted into a computer program, where each line of the program calls a task-specific audio generation model or computational operation function. The computer program is then executed to obtain a compositional and interpretable solution for audio creation. Experimental results suggest that WavJourney is capable of synthesizing realistic audio aligned with textually-described semantic, spatial and temporal conditions, achieving state-of-the-art results on text-to-audio generation benchmarks. Additionally, we introduce a new multi-genre story benchmark. Subjective evaluations demonstrate the potential of WavJourney in crafting engaging storytelling audio content from text. We further demonstrate that WavJourney can facilitate human-machine co-creation in multi-round dialogues. To foster future research, the code and synthesized audio are available at: https://audio-agi.github.io/WavJourney_demopage/.

Shizun Wang, Weihong Zeng, Xu Wang et al.

The creation of a parameterized stylized character involves careful selection of numerous parameters, also known as the "avatar vectors" that can be interpreted by the avatar engine. Existing unsupervised avatar vector estimation methods that auto-create avatars for users, however, often fail to work because of the domain gap between realistic faces and stylized avatar images. To this end, we propose SwiftAvatar, a novel avatar auto-creation framework that is evidently superior to previous works. SwiftAvatar introduces dual-domain generators to create pairs of realistic faces and avatar images using shared latent codes. The latent codes can then be bridged with the avatar vectors as pairs, by performing GAN inversion on the avatar images rendered from the engine using avatar vectors. Through this way, we are able to synthesize paired data in high-quality as many as possible, consisting of avatar vectors and their corresponding realistic faces. We also propose semantic augmentation to improve the diversity of synthesis. Finally, a light-weight avatar vector estimator is trained on the synthetic pairs to implement efficient auto-creation. Our experiments demonstrate the effectiveness and efficiency of SwiftAvatar on two different avatar engines. The superiority and advantageous flexibility of SwiftAvatar are also verified in both subjective and objective evaluations.

Yi Yuan, Haohe Liu, Jinhua Liang et al.

Deep neural networks have recently achieved breakthroughs in sound generation. Despite the outstanding sample quality, current sound generation models face issues on small-scale datasets (e.g., overfitting), significantly limiting performance. In this paper, we make the first attempt to investigate the benefits of pre-training on sound generation with AudioLDM, the cutting-edge model for audio generation, as the backbone. Our study demonstrates the advantages of the pre-trained AudioLDM, especially in data-scarcity scenarios. In addition, the baselines and evaluation protocol for sound generation systems are not consistent enough to compare different studies directly. Aiming to facilitate further study on sound generation tasks, we benchmark the sound generation task on various frequently-used datasets. We hope our results on transfer learning and benchmarks can provide references for further research on conditional sound generation.

Junqi Zhao, Xubo Liu, Jinzheng Zhao et al.

Universal sound separation (USS) is a task of separating mixtures of arbitrary sound sources. Typically, universal separation models are trained from scratch in a supervised manner, using labeled data. Self-supervised learning (SSL) is an emerging deep learning approach that leverages unlabeled data to obtain task-agnostic representations, which can benefit many downstream tasks. In this paper, we propose integrating a self-supervised pre-trained model, namely the audio masked autoencoder (A-MAE), into a universal sound separation system to enhance its separation performance. We employ two strategies to utilize SSL embeddings: freezing or updating the parameters of A-MAE during fine-tuning. The SSL embeddings are concatenated with the short-time Fourier transform (STFT) to serve as input features for the separation model. We evaluate our methods on the AudioSet dataset, and the experimental results indicate that the proposed methods successfully enhance the separation performance of a state-of-the-art ResUNet-based USS model.

Yi Yuan, Haohe Liu, Xubo Liu et al.

Foley sound generation aims to synthesise the background sound for multimedia content. Previous models usually employ a large development set with labels as input (e.g., single numbers or one-hot vector). In this work, we propose a diffusion model based system for Foley sound generation with text conditions. To alleviate the data scarcity issue, our model is initially pre-trained with large-scale datasets and fine-tuned to this task via transfer learning using the contrastive language-audio pertaining (CLAP) technique. We have observed that the feature embedding extracted by the text encoder can significantly affect the performance of the generation model. Hence, we introduce a trainable layer after the encoder to improve the text embedding produced by the encoder. In addition, we further refine the generated waveform by generating multiple candidate audio clips simultaneously and selecting the best one, which is determined in terms of the similarity score between the embedding of the candidate clips and the embedding of the target text label. Using the proposed method, our system ranks ${1}^{st}$ among the systems submitted to DCASE Challenge 2023 Task 7. The results of the ablation studies illustrate that the proposed techniques significantly improve sound generation performance. The codes for implementing the proposed system are available online.

Zhiqian Lin, Jiangke Lin, Lincheng Li et al.

Recent works based on convolutional encoder-decoder architecture and 3DMM parameterization have shown great potential for canonical view reconstruction from a single input image. Conventional CNN architectures benefit from exploiting the spatial correspondence between the input and output pixels. However, in 3D face reconstruction, the spatial misalignment between the input image (e.g. face) and the canonical/UV output makes the feature encoding-decoding process quite challenging. In this paper, to tackle this problem, we propose a new network architecture, namely the Affine Convolution Networks, which enables CNN based approaches to handle spatially non-corresponding input and output images and maintain high-fidelity quality output at the same time. In our method, an affine transformation matrix is learned from the affine convolution layer for each spatial location of the feature maps. In addition, we represent 3D human heads in UV space with multiple components, including diffuse maps for texture representation, position maps for geometry representation, and light maps for recovering more complex lighting conditions in the real world. All the components can be trained without any manual annotations. Our method is parametric-free and can generate high-quality UV maps at resolution of 512 x 512 pixels, while previous approaches normally generate 256 x 256 pixels or smaller. Our code will be released once the paper got accepted.

Zhucun Xue, Jiangning Zhang, Juntao Jiang et al.

The rapid expansion of research across machine learning, vision, and language has produced a volume of publications that is increasingly difficult to synthesize. Traditional bibliometric tools rely mainly on metadata and offer limited visibility into the semantic content of papers, making it hard to track how research themes evolve over time or how different areas influence one another. To obtain a clearer picture of recent developments, we compile a unified corpus of more than 100,000 papers from 22 major conferences between 2020 and 2025 and construct a multidimensional profiling pipeline to organize and analyze their textual content. By combining topic clustering, LLM-assisted parsing, and structured retrieval, we derive a comprehensive representation of research activity that supports the study of topic lifecycles, methodological transitions, dataset and model usage patterns, and institutional research directions. Our analysis highlights several notable shifts, including the growth of safety, multimodal reasoning, and agent-oriented studies, as well as the gradual stabilization of areas such as neural machine translation and graph-based methods. These findings provide an evidence-based view of how AI research is evolving and offer a resource for understanding broader trends and identifying emerging directions. Code and dataset: https://github.com/xzc-zju/Profiling_Scientific_Literature

Ruobing Zheng, Tianqi Li, Jianing Li et al.

Reasoning post-training improves Large Language Models (LLMs) on complex tasks such as mathematics and coding, but its benefits across diverse multimodal tasks remains uncertain. The trend of releasing parallel "Instruct" and "Thinking" models by leading teams is both resource-intensive and user-unfriendly. Prior work finds that the gains from reasoning training are influenced by multiple factors, such as base model capabilities, task characteristics, and Chain-of-Thought (CoT) data quality. However, principled criteria for determining when reasoning post-training is beneficial and which data should support it are still lacking. In this paper, we propose Dual Tuning, a reasoning efficacy-driven data curation framework for multimodal LLMs training. Given a target task and a base model, Dual Tuning jointly evaluates whether the training data is beneficial and whether reasoning training with current CoT content yields positive gains over non-reasoning alternatives. We apply Dual Tuning across spatial, mathematical, and multi-disciplinary tasks, and further analyze how reinforcement learning and thinking patterns affect reasoning efficacy. The Dual Tuning results guide data curation by identifying data that benefit reasoning training, data better suited to direct-answer training, and data that are detrimental under both training modes. Our work provides quantitative criteria for selecting appropriate training data and matching post-training strategies.

Haohe Liu, Xuenan Xu, Yi Yuan et al.

Large language models (LLMs) have significantly advanced audio processing through audio codecs that convert audio into discrete tokens, enabling the application of language modelling techniques to audio data. However, traditional codecs often operate at high bitrates or within narrow domains such as speech and lack the semantic clues required for efficient language modelling. Addressing these challenges, we introduce SemantiCodec, a novel codec designed to compress audio into fewer than a hundred tokens per second across diverse audio types, including speech, general sound, and music, without compromising quality. SemantiCodec features a dual-encoder architecture: a semantic encoder using a self-supervised pre-trained Audio Masked Autoencoder (AudioMAE), discretized using k-means clustering on extensive audio data, and an acoustic encoder to capture the remaining details. The semantic and acoustic encoder outputs are used to reconstruct audio via a diffusion-model-based decoder. SemantiCodec is presented in three variants with token rates of 25, 50, and 100 per second, supporting a range of ultra-low bit rates between 0.31 kbps and 1.40 kbps. Experimental results demonstrate that SemantiCodec significantly outperforms the state-of-the-art Descript codec on reconstruction quality. Our results also suggest that SemantiCodec contains significantly richer semantic information than all evaluated state-of-the-art audio codecs, even at significantly lower bitrates. Our code and demos are available at https://haoheliu.github.io/SemantiCodec/.

Inclusion AI, Biao Gong, Cheng Zou et al.

We propose Ming-Omni, a unified multimodal model capable of processing images, text, audio, and video, while demonstrating strong proficiency in both speech and image generation. Ming-Omni employs dedicated encoders to extract tokens from different modalities, which are then processed by Ling, an MoE architecture equipped with newly proposed modality-specific routers. This design enables a single model to efficiently process and fuse multimodal inputs within a unified framework, thereby facilitating diverse tasks without requiring separate models, task-specific fine-tuning, or structural redesign. Importantly, Ming-Omni extends beyond conventional multimodal models by supporting audio and image generation. This is achieved through the integration of an advanced audio decoder for natural-sounding speech and Ming-Lite-Uni for high-quality image generation, which also allow the model to engage in context-aware chatting, perform text-to-speech conversion, and conduct versatile image editing. Our experimental results showcase Ming-Omni offers a powerful solution for unified perception and generation across all modalities. Notably, our proposed Ming-Omni is the first open-source model we are aware of to match GPT-4o in modality support, and we release all code and model weights to encourage further research and development in the community.

Canxiang Yan, Chunxiang Jin, Dawei Huang et al.

Existing speech models suffer from competing requirements on token representations by understanding and generation tasks. This discrepancy in representation prevents speech language models from performing instruction-based free-form editing. To solve this challenge, we introduce a novel framework that unifies speech understanding, generation, and editing. The core of our unified model is a unified continuous speech tokenizer MingTok-Audio, the first continuous tokenizer to effectively integrate semantic and acoustic features, which makes it suitable for both understanding and generation tasks. Based on this unified continuous audio tokenizer, we developed the speech language model Ming-UniAudio, which achieved a balance between generation and understanding capabilities. Ming-UniAudio sets new state-of-the-art (SOTA) records on 8 out of 12 metrics on the ContextASR benchmark. Notably, for Chinese voice cloning, it achieves a highly competitive Seed-TTS-WER of 0.95. Leveraging this foundational model, we further trained a dedicated speech editing model Ming-UniAudio-Edit, the first speech language model that enables universal, free-form speech editing guided solely by natural language instructions, handling both semantic and acoustic modifications without timestamp condition. To rigorously assess the editing capability and establish a foundation for future research, we introduce Ming-Freeform-Audio-Edit, the first comprehensive benchmark tailored for instruction-based free-form speech editing, featuring diverse scenarios and evaluation dimensions spanning semantic correctness, acoustic quality, and instruction alignment. We open-sourced the continuous audio tokenizer, the unified foundational model, and the free-form instruction-based editing model to facilitate the development of unified audio understanding, generation, and manipulation.

Lei Li, Ze Zhao, Meng Li et al.

Document parsing, as a fundamental yet crucial vision task, is being revolutionized by vision-language models (VLMs). However, the autoregressive (AR) decoding inherent to VLMs creates a significant bottleneck, severely limiting parsing speed. In this paper, we propose Parallel-Token Prediction (PTP), a plugable, model-agnostic and simple-yet-effective method that enables VLMs to generate multiple future tokens in parallel with improved sample efficiency. Specifically, we insert some learnable tokens into the input sequence and design corresponding training objectives to equip the model with parallel decoding capabilities for document parsing. Furthermore, to support effective training, we develop a comprehensive data generation pipeline that efficiently produces large-scale, high-quality document parsing training data for VLMs. Extensive experiments on OmniDocBench and olmOCR-bench demonstrate that our method not only significantly improves decoding speed (1.6x-2.2x) but also reduces model hallucinations and exhibits strong generalization abilities.

Zhengxia Zou, Tianyang Shi, Yi Yuan et al.

An autostereogram, a.k.a. magic eye image, is a single-image stereogram that can create visual illusions of 3D scenes from 2D textures. This paper studies an interesting question that whether a deep CNN can be trained to recover the depth behind an autostereogram and understand its content. The key to the autostereogram magic lies in the stereopsis - to solve such a problem, a model has to learn to discover and estimate disparity from the quasi-periodic textures. We show that deep CNNs embedded with disparity convolution, a novel convolutional layer proposed in this paper that simulates stereopsis and encodes disparity, can nicely solve such a problem after being sufficiently trained on a large 3D object dataset in a self-supervised fashion. We refer to our method as ``NeuralMagicEye''. Experiments show that our method can accurately recover the depth behind autostereograms with rich details and gradient smoothness. Experiments also show the completely different working mechanisms for autostereogram perception between neural networks and human eyes. We hope this research can help people with visual impairments and those who have trouble viewing autostereograms. Our code is available at \url{https://jiupinjia.github.io/neuralmagiceye/}.

Xiaoyang Lyu, Liang Liu, Mengmeng Wang et al.

Self-supervised learning shows great potential in monoculardepth estimation, using image sequences as the only source ofsupervision. Although people try to use the high-resolutionimage for depth estimation, the accuracy of prediction hasnot been significantly improved. In this work, we find thecore reason comes from the inaccurate depth estimation inlarge gradient regions, making the bilinear interpolation er-ror gradually disappear as the resolution increases. To obtainmore accurate depth estimation in large gradient regions, itis necessary to obtain high-resolution features with spatialand semantic information. Therefore, we present an improvedDepthNet, HR-Depth, with two effective strategies: (1) re-design the skip-connection in DepthNet to get better high-resolution features and (2) propose feature fusion Squeeze-and-Excitation(fSE) module to fuse feature more efficiently.Using Resnet-18 as the encoder, HR-Depth surpasses all pre-vious state-of-the-art(SoTA) methods with the least param-eters at both high and low resolution. Moreover, previousstate-of-the-art methods are based on fairly complex and deepnetworks with a mass of parameters which limits their realapplications. Thus we also construct a lightweight networkwhich uses MobileNetV3 as encoder. Experiments show thatthe lightweight network can perform on par with many largemodels like Monodepth2 at high-resolution with only20%parameters. All codes and models will be available at https://github.com/shawLyu/HR-Depth.

Yi Yuan, Zhuo Chen, Xubo Liu et al.

Contrastive language-audio pretraining~(CLAP) has been developed to align the representations of audio and language, achieving remarkable performance in retrieval and classification tasks. However, current CLAP struggles to capture temporal information within audio and text features, presenting substantial limitations for tasks such as audio retrieval and generation. To address this gap, we introduce T-CLAP, a temporal-enhanced CLAP model. We use Large Language Models~(LLMs) and mixed-up strategies to generate temporal-contrastive captions for audio clips from extensive audio-text datasets. Subsequently, a new temporal-focused contrastive loss is designed to fine-tune the CLAP model by incorporating these synthetic data. We conduct comprehensive experiments and analysis in multiple downstream tasks. T-CLAP shows improved capability in capturing the temporal relationship of sound events and outperforms state-of-the-art models by a significant margin.

Ang Li, Qiugen Xiao, Peng Cao et al.

Reinforcement Learning from AI Feedback (RLAIF) has the advantages of shorter annotation cycles and lower costs over Reinforcement Learning from Human Feedback (RLHF), making it highly efficient during the rapid strategy iteration periods of large language model (LLM) training. Using ChatGPT as a labeler to provide feedback on open-domain prompts in RLAIF training, we observe an increase in human evaluators' preference win ratio for model responses, but a decrease in evaluators' satisfaction rate. Analysis suggests that the decrease in satisfaction rate is mainly due to some responses becoming less helpful, particularly in terms of correctness and truthfulness, highlighting practical limitations of basic RLAIF. In this paper, we propose Hybrid Reinforcement Learning from AI Feedback (HRLAIF). This method enhances the accuracy of AI annotations for responses, making the model's helpfulness more robust in training process. Additionally, it employs AI for Red Teaming, further improving the model's harmlessness. Human evaluation results show that HRLAIF inherits the ability of RLAIF to enhance human preference for outcomes at a low cost while also improving the satisfaction rate of responses. Compared to the policy model before Reinforcement Learning (RL), it achieves an increase of 2.08\% in satisfaction rate, effectively addressing the issue of a decrease of 4.58\% in satisfaction rate after basic RLAIF.

Inclusion AI, Fudong Wang, Jiajia Liu et al.

Recent advancements in Multimodal Large Language Models (MLLMs), particularly through Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced their reasoning abilities. However, a critical gap persists: these models struggle with dynamic spatial interactions, a capability essential for real-world applications. To bridge this gap, we introduce M2-Reasoning-7B, a model designed to excel in both general and spatial reasoning. Our approach integrates two key innovations: (1) a novel data pipeline that generates 294.2K high-quality data samples (168K for cold-start fine-tuning and 126.2K for RLVR), which feature logically coherent reasoning trajectories and have undergone comprehensive assessment; and (2) a dynamic multi-task training strategy with step-wise optimization to mitigate conflicts between data, and task-specific rewards for delivering tailored incentive signals. This combination of curated data and advanced training allows M2-Reasoning-7B to set a new state-of-the-art (SOTA) across 8 benchmarks, showcasing superior performance in both general and spatial reasoning domains.

Yi Yuan, Xuhong Wang, Shanzhe Lei

As agent-based systems continue to evolve, deep research agents are capable of automatically generating research-style reports across diverse domains. While these agents promise to streamline information synthesis and knowledge exploration, existing evaluation frameworks-typically based on subjective dimensions-fail to capture a critical aspect of report quality: trustworthiness. In open-ended research scenarios where ground-truth answers are unavailable, current evaluation methods cannot effectively measure the epistemic confidence of generated content, making calibration difficult and leaving users susceptible to misleading or hallucinated information. To address this limitation, we propose a novel deep research agent that incorporates progressive confidence estimation and calibration within the report generation pipeline. Our system leverages a deliberative search model, featuring deep retrieval and multi-hop reasoning to ground outputs in verifiable evidence while assigning confidence scores to individual claims. Combined with a carefully designed workflow, this approach produces trustworthy reports with enhanced transparency. Experimental results and case studies demonstrate that our method substantially improves interpretability and significantly increases user trust.

Inclusion AI, Bowen Ma, Cheng Zou et al.

We propose Ming-Flash-Omni, an upgraded version of Ming-Omni, built upon a sparser Mixture-of-Experts (MoE) variant of Ling-Flash-2.0 with 100 billion total parameters, of which only 6.1 billion are active per token. This architecture enables highly efficient scaling (dramatically improving computational efficiency while significantly expanding model capacity) and empowers stronger unified multimodal intelligence across vision, speech, and language, representing a key step toward Artificial General Intelligence (AGI). Compared to its predecessor, the upgraded version exhibits substantial improvements across multimodal understanding and generation. We significantly advance speech recognition capabilities, achieving state-of-the-art performance in contextual ASR and highly competitive results in dialect-aware ASR. In image generation, Ming-Flash-Omni introduces high-fidelity text rendering and demonstrates marked gains in scene consistency and identity preservation during image editing. Furthermore, Ming-Flash-Omni introduces generative segmentation, a capability that not only achieves strong standalone segmentation performance but also enhances spatial control in image generation and improves editing consistency. Notably, Ming-Flash-Omni achieves state-of-the-art results in text-to-image generation and generative segmentation, and sets new records on all 12 contextual ASR benchmarks, all within a single unified architecture.

Zheng Qin, Ruobing Zheng, Yabing Wang et al.

While Multimodal Large Language Models (MLLMs) show immense promise for achieving truly human-like interactions, progress is hindered by the lack of fine-grained evaluation frameworks for human-centered scenarios, encompassing both the understanding of complex human intentions and the provision of empathetic, context-aware responses. Here we introduce HumanSense, a comprehensive benchmark designed to evaluate the human-centered perception and interaction capabilities of MLLMs, with a particular focus on deep understanding of extended multimodal contexts and the formulation of rational feedback. Our evaluation reveals that leading MLLMs still have considerable room for improvement, particularly for advanced interaction-oriented tasks. Supplementing visual input with audio and text information yields substantial improvements, and Omni-modal models show advantages on these tasks.Furthermore, grounded in the observation that appropriate feedback stems from a contextual analysis of the interlocutor's needs and emotions, we posit that reasoning ability serves as the key to unlocking it. We devise a multi-stage, modality-progressive reinforcement learning approach, resulting in HumanSense-Omni-Reasoning, which substantially enhances performance on higher-level understanding and interactive tasks. Additionally, we observe that successful reasoning processes appear to exhibit consistent thought patterns. By designing corresponding prompts, we also enhance the performance of non-reasoning models in a training-free manner.Project page: \textcolor{brightpink}{https://digital-avatar.github.io/ai/HumanSense/}

Liting Gao, Yi Yuan, Yaru Chen et al.

Diffusion models have shown remarkable progress in text-to-audio generation. However, text-guided audio editing remains in its early stages. This task focuses on modifying the target content within an audio signal while preserving the rest, thus demanding precise localization and faithful editing according to the text prompt. Existing training-based and zero-shot methods that rely on full-caption or costly optimization often struggle with complex editing or lack practicality. In this work, we propose a novel end-to-end efficient rectified flow matching-based diffusion framework for audio editing, and construct a dataset featuring overlapping multi-event audio to support training and benchmarking in complex scenarios. Experiments show that our model achieves faithful semantic alignment without requiring auxiliary captions or masks, while maintaining competitive editing quality across metrics.

Yi Yuan, Xubo Liu, Haohe Liu et al.

With the development of large-scale diffusion-based and language-modeling-based generative models, impressive progress has been achieved in text-to-audio generation. Despite producing high-quality outputs, existing text-to-audio models mainly aim to generate semantically aligned sound and fall short on precisely controlling fine-grained acoustic characteristics of specific sounds. As a result, users that need specific sound content may find it challenging to generate the desired audio clips. In this paper, we present DreamAudio for customized text-to-audio generation (CTTA). Specifically, we introduce a new framework that is designed to enable the model to identify auditory information from user-provided reference concepts for audio generation. Given a few reference audio samples containing personalized audio events, our system can generate new audio samples that include these specific events. In addition, two types of datasets are developed for training and testing the customized systems. The experiments show that the proposed model, DreamAudio, generates audio samples that are highly consistent with the customized audio features and aligned well with the input text prompts. Furthermore, DreamAudio offers comparable performance in general text-to-audio tasks. We also provide a human-involved dataset containing audio events from real-world CTTA cases as the benchmark for customized generation tasks.

Yi Yuan, Joseph Van Duyn, Runze Yan et al. · stanford

Electrocardiogram (ECG) analysis is foundational for cardiovascular disease diagnosis, yet the performance of deep learning models is often constrained by limited access to annotated data. Self-supervised contrastive learning has emerged as a powerful approach for learning robust ECG representations from unlabeled signals. However, most existing methods generate only pairwise augmented views and fail to leverage the rich temporal structure of ECG recordings. In this work, we present a poly-window contrastive learning framework. We extract multiple temporal windows from each ECG instance to construct positive pairs and maximize their agreement via statistics. Inspired by the principle of slow feature analysis, our approach explicitly encourages the model to learn temporally invariant and physiologically meaningful features that persist across time. We validate our approach through extensive experiments and ablation studies on the PTB-XL dataset. Our results demonstrate that poly-window contrastive learning consistently outperforms conventional two-view methods in multi-label superclass classification, achieving higher AUROC (0.891 vs. 0.888) and F1 scores (0.680 vs. 0.679) while requiring up to four times fewer pre-training epochs (32 vs. 128) and 14.8% in total wall clock pre-training time reduction. Despite processing multiple windows per sample, we achieve a significant reduction in the number of training epochs and total computation time, making our method practical for training foundational models. Through extensive ablations, we identify optimal design choices and demonstrate robustness across various hyperparameters. These findings establish poly-window contrastive learning as a highly efficient and scalable paradigm for automated ECG analysis and provide a promising general framework for self-supervised representation learning in biomedical time-series data.

Guangming Yao, Hongzhi Wu, Yi Yuan et al.

In this paper, we present a novel double diffusion based neural radiance field, dubbed DD-NeRF, to reconstruct human body geometry and render the human body appearance in novel views from a sparse set of images. We first propose a double diffusion mechanism to achieve expressive representations of input images by fully exploiting human body priors and image appearance details at two levels. At the coarse level, we first model the coarse human body poses and shapes via an unclothed 3D deformable vertex model as guidance. At the fine level, we present a multi-view sampling network to capture subtle geometric deformations and image detailed appearances, such as clothing and hair, from multiple input views. Considering the sparsity of the two level features, we diffuse them into feature volumes in the canonical space to construct neural radiance fields. Then, we present a signed distance function (SDF) regression network to construct body surfaces from the diffused features. Thanks to our double diffused representations, our method can even synthesize novel views of unseen subjects. Experiments on various datasets demonstrate that our approach outperforms the state-of-the-art in both geometric reconstruction and novel view synthesis.

Qi Wen, Shuang Li, Bingfeng Han et al.

Chinese character style transfer is a very challenging problem because of the complexity of the glyph shapes or underlying structures and large numbers of existed characters, when comparing with English letters. Moreover, the handwriting of calligraphy masters has a more irregular stroke and is difficult to obtain in real-world scenarios. Recently, several GAN-based methods have been proposed for font synthesis, but some of them require numerous reference data and the other part of them have cumbersome preprocessing steps to divide the character into different parts to be learned and transferred separately. In this paper, we propose a simple but powerful end-to-end Chinese calligraphy font generation framework ZiGAN, which does not require any manual operation or redundant preprocessing to generate fine-grained target-style characters with few-shot references. To be specific, a few paired samples from different character styles are leveraged to attain a fine-grained correlation between structures underlying different glyphs. To capture valuable style knowledge in target and strengthen the coarse-grained understanding of character content, we utilize multiple unpaired samples to align the feature distributions belonging to different character styles. By doing so, only a few target Chinese calligraphy characters are needed to generated expected style transferred characters. Experiments demonstrate that our method has a state-of-the-art generalization ability in few-shot Chinese character style transfer.

Jianbiao Mei, Mengmeng Wang, Yeneng Lin et al.

Recently, Space-Time Memory Network (STM) based methods have achieved state-of-the-art performance in semi-supervised video object segmentation (VOS). A crucial problem in this task is how to model the dependency both among different frames and inside every frame. However, most of these methods neglect the spatial relationships (inside each frame) and do not make full use of the temporal relationships (among different frames). In this paper, we propose a new transformer-based framework, termed TransVOS, introducing a vision transformer to fully exploit and model both the temporal and spatial relationships. Moreover, most STM-based approaches employ two separate encoders to extract features of two significant inputs, i.e., reference sets (history frames with predicted masks) and query frame (current frame), respectively, increasing the models' parameters and complexity. To slim the popular two-encoder pipeline while keeping the effectiveness, we design a single two-path feature extractor to encode the above two inputs in a unified way. Extensive experiments demonstrate the superiority of our TransVOS over state-of-the-art methods on both DAVIS and YouTube-VOS datasets.

Yinglin Duan, Tianyang Shi, Zhengxia Zou et al.

Motion completion is a challenging and long-discussed problem, which is of great significance in film and game applications. For different motion completion scenarios (in-betweening, in-filling, and blending), most previous methods deal with the completion problems with case-by-case designs. In this work, we propose a simple but effective method to solve multiple motion completion problems under a unified framework and achieves a new state of the art accuracy under multiple evaluation settings. Inspired by the recent great success of attention-based models, we consider the completion as a sequence to sequence prediction problem. Our method consists of two modules - a standard transformer encoder with self-attention that learns long-range dependencies of input motions, and a trainable mixture embedding module that models temporal information and discriminates key-frames. Our method can run in a non-autoregressive manner and predict multiple missing frames within a single forward propagation in real time. We finally show the effectiveness of our method in music-dance applications.

Lijuan Liu, Yin Yang, Yi Yuan et al.

In this paper, we propose an effective global relation learning algorithm to recommend an appropriate location of a building unit for in-game customization of residential home complex. Given a construction layout, we propose a visual context-aware graph generation network that learns the implicit global relations among the scene components and infers the location of a new building unit. The proposed network takes as input the scene graph and the corresponding top-view depth image. It provides the location recommendations for a newly-added building units by learning an auto-regressive edge distribution conditioned on existing scenes. We also introduce a global graph-image matching loss to enhance the awareness of essential geometry semantics of the site. Qualitative and quantitative experiments demonstrate that the recommended location well reflects the implicit spatial rules of components in the residential estates, and it is instructive and practical to locate the building units in the 3D scene of the complex construction.

Guangming Yao, Yi Yuan, Tianjia Shao et al.

The paper proposes a novel generative adversarial network for one-shot face reenactment, which can animate a single face image to a different pose-and-expression (provided by a driving image) while keeping its original appearance. The core of our network is a novel mechanism called appearance adaptive normalization, which can effectively integrate the appearance information from the input image into our face generator by modulating the feature maps of the generator using the learned adaptive parameters. Furthermore, we specially design a local net to reenact the local facial components (i.e., eyes, nose and mouth) first, which is a much easier task for the network to learn and can in turn provide explicit anchors to guide our face generator to learn the global appearance and pose-and-expression. Extensive quantitative and qualitative experiments demonstrate the significant efficacy of our model compared with prior one-shot methods.

Jilin Tang, Yi Yuan, Tianjia Shao et al.

In this paper we tackle the problem of pose guided person image generation, which aims to transfer a person image from the source pose to a novel target pose while maintaining the source appearance. Given the inefficiency of standard CNNs in handling large spatial transformation, we propose a structure-aware flow based method for high-quality person image generation. Specifically, instead of learning the complex overall pose changes of human body, we decompose the human body into different semantic parts (e.g., head, torso, and legs) and apply different networks to predict the flow fields for these parts separately. Moreover, we carefully design the network modules to effectively capture the local and global semantic correlations of features within and among the human parts respectively. Extensive experimental results show that our method can generate high-quality results under large pose discrepancy and outperforms state-of-the-art methods in both qualitative and quantitative comparisons.

Jiangke Lin, Yi Yuan, Zhengxia Zou

Many deep learning based 3D face reconstruction methods have been proposed recently, however, few of them have applications in games. Current game character customization systems either require players to manually adjust considerable face attributes to obtain the desired face, or have limited freedom of facial shape and texture. In this paper, we propose an automatic character face creation method that predicts both facial shape and texture from a single portrait, and it can be integrated into most existing 3D games. Although 3D Morphable Face Model (3DMM) based methods can restore accurate 3D faces from single images, the topology of 3DMM mesh is different from the meshes used in most games. To acquire fidelity texture, existing methods require a large amount of face texture data for training, while building such datasets is time-consuming and laborious. Besides, such a dataset collected under laboratory conditions may not generalized well to in-the-wild situations. To tackle these problems, we propose 1) a low-cost facial texture acquisition method, 2) a shape transfer algorithm that can transform the shape of a 3DMM mesh to games, and 3) a new pipeline for training 3D game face reconstruction networks. The proposed method not only can produce detailed and vivid game characters similar to the input portrait, but can also eliminate the influence of lighting and occlusions. Experiments show that our method outperforms state-of-the-art methods used in games.

Zhengxia Zou, Tianyang Shi, Shuang Qiu et al.

This paper proposes an image-to-painting translation method that generates vivid and realistic painting artworks with controllable styles. Different from previous image-to-image translation methods that formulate the translation as pixel-wise prediction, we deal with such an artistic creation process in a vectorized environment and produce a sequence of physically meaningful stroke parameters that can be further used for rendering. Since a typical vector render is not differentiable, we design a novel neural renderer which imitates the behavior of the vector renderer and then frame the stroke prediction as a parameter searching process that maximizes the similarity between the input and the rendering output. We explored the zero-gradient problem on parameter searching and propose to solve this problem from an optimal transportation perspective. We also show that previous neural renderers have a parameter coupling problem and we re-design the rendering network with a rasterization network and a shading network that better handles the disentanglement of shape and color. Experiments show that the paintings generated by our method have a high degree of fidelity in both global appearance and local textures. Our method can be also jointly optimized with neural style transfer that further transfers visual style from other images. Our code and animated results are available at \url{https://jiupinjia.github.io/neuralpainter/}.

Yinglin Duan, Tianyang Shi, Zhengxia Zou et al.

Music-to-dance translation is a brand-new and powerful feature in recent role-playing games. Players can now let their characters dance along with specified music clips and even generate fan-made dance videos. Previous works of this topic consider music-to-dance as a supervised motion generation problem based on time-series data. However, these methods suffer from limited training data pairs and the degradation of movements. This paper provides a new perspective for this task where we re-formulate the translation problem as a piece-wise dance phrase retrieval problem based on the choreography theory. With such a design, players are allowed to further edit the dance movements on top of our generation while other regression based methods ignore such user interactivity. Considering that the dance motion capture is an expensive and time-consuming procedure which requires the assistance of professional dancers, we train our method under a semi-supervised learning framework with a large unlabeled dataset (20x than labeled data) collected. A co-ascent mechanism is introduced to improve the robustness of our network. Using this unlabeled dataset, we also introduce self-supervised pre-training so that the translator can understand the melody, rhythm, and other components of music phrases. We show that the pre-training significantly improves the translation accuracy than that of training from scratch. Experimental results suggest that our method not only generalizes well over various styles of music but also succeeds in expert-level choreography for game players.

Wenheng Chen, He Wang, Yi Yuan et al.

Human motion modelling is crucial in many areas such as computer graphics, vision and virtual reality. Acquiring high-quality skeletal motions is difficult due to the need for specialized equipment and laborious manual post-posting, which necessitates maximizing the use of existing data to synthesize new data. However, it is a challenge due to the intrinsic motion stochasticity of human motion dynamics, manifested in the short and long terms. In the short term, there is strong randomness within a couple frames, e.g. one frame followed by multiple possible frames leading to different motion styles; while in the long term, there are non-deterministic action transitions. In this paper, we present Dynamic Future Net, a new deep learning model where we explicitly focuses on the aforementioned motion stochasticity by constructing a generative model with non-trivial modelling capacity in temporal stochasticity. Given limited amounts of data, our model can generate a large number of high-quality motions with arbitrary duration, and visually-convincing variations in both space and time. We evaluate our model on a wide range of motions and compare it with the state-of-the-art methods. Both qualitative and quantitative results show the superiority of our method, for its robustness, versatility and high-quality.

Xinhui Song, Tianyang Shi, Zunlei Feng et al.

Facial action unit (AU) intensity is an index to describe all visually discernible facial movements. Most existing methods learn intensity estimator with limited AU data, while they lack generalization ability out of the dataset. In this paper, we present a framework to predict the facial parameters (including identity parameters and AU parameters) based on a bone-driven face model (BDFM) under different views. The proposed framework consists of a feature extractor, a generator, and a facial parameter regressor. The regressor can fit the physical meaning parameters of the BDFM from a single face image with the help of the generator, which maps the facial parameters to the game-face images as a differentiable renderer. Besides, identity loss, loopback loss, and adversarial loss can improve the regressive results. Quantitative evaluations are performed on two public databases BP4D and DISFA, which demonstrates that the proposed method can achieve comparable or better performance than the state-of-the-art methods. What's more, the qualitative results also demonstrate the validity of our method in the wild.

Guangming Yao, Yi Yuan, Tianjia Shao et al.

Face reenactment aims to animate a source face image to a different pose and expression provided by a driving image. Existing approaches are either designed for a specific identity, or suffer from the identity preservation problem in the one-shot or few-shot scenarios. In this paper, we introduce a method for one-shot face reenactment, which uses the reconstructed 3D meshes (i.e., the source mesh and driving mesh) as guidance to learn the optical flow needed for the reenacted face synthesis. Technically, we explicitly exclude the driving face's identity information in the reconstructed driving mesh. In this way, our network can focus on the motion estimation for the source face without the interference of driving face shape. We propose a motion net to learn the face motion, which is an asymmetric autoencoder. The encoder is a graph convolutional network (GCN) that learns a latent motion vector from the meshes, and the decoder serves to produce an optical flow image from the latent vector with CNNs. Compared to previous methods using sparse keypoints to guide the optical flow learning, our motion net learns the optical flow directly from 3D dense meshes, which provide the detailed shape and pose information for the optical flow, so it can achieve more accurate expression and pose on the reenacted face. Extensive experiments show that our method can generate high-quality results and outperforms state-of-the-art methods in both qualitative and quantitative comparisons.

Tianyang Shi, Zhengxia Zou, Xinhui Song et al.

Game character customization is one of the core features of many recent Role-Playing Games (RPGs), where players can edit the appearance of their in-game characters with their preferences. This paper studies the problem of automatically creating in-game characters with a single photo. In recent literature on this topic, neural networks are introduced to make game engine differentiable and the self-supervised learning is used to predict facial customization parameters. However, in previous methods, the expression parameters and facial identity parameters are highly coupled with each other, making it difficult to model the intrinsic facial features of the character. Besides, the neural network based renderer used in previous methods is also difficult to be extended to multi-view rendering cases. In this paper, considering the above problems, we propose a novel method named "PokerFace-GAN" for neutral face game character auto-creation. We first build a differentiable character renderer which is more flexible than the previous methods in multi-view rendering cases. We then take advantage of the adversarial training to effectively disentangle the expression parameters from the identity parameters and thus generate player-preferred neutral face (expression-less) characters. Since all components of our method are differentiable, our method can be easily trained under a multi-task self-supervised learning paradigm. Experiment results show that our method can generate vivid neutral face game characters that are highly similar to the input photos. The effectiveness of our method is verified by comparison results and ablation studies.

Tianyang Shi, Zhengxia Zou, Yi Yuan et al.

With the rapid development of Role-Playing Games (RPGs), players are now allowed to edit the facial appearance of their in-game characters with their preferences rather than using default templates. This paper proposes a game character auto-creation framework that generates in-game characters according to a player's input face photo. Different from the previous methods that are designed based on neural style transfer or monocular 3D face reconstruction, we re-formulate the character auto-creation process in a different point of view: by predicting a large set of physically meaningful facial parameters under a self-supervised learning paradigm. Instead of updating facial parameters iteratively at the input end of the renderer as suggested by previous methods, which are time-consuming, we introduce a facial parameter translator so that the creation can be done efficiently through a single forward propagation from the face embeddings to parameters, with a considerable 1000x computational speedup. Despite its high efficiency, the interactivity is preserved in our method where users are allowed to optionally fine-tune the facial parameters on our creation according to their needs. Our approach also shows better robustness than previous methods, especially for those photos with head-pose variance. Comparison results and ablation analysis on seven public face verification datasets suggest the effectiveness of our method.

Yifan Xu, Tianqi Fan, Yi Yuan et al.

Deep implicit field regression methods are effective for 3D reconstruction from single-view images. However, the impact of different sampling patterns on the reconstruction quality is not well-understood. In this work, we first study the effect of point set discrepancy on the network training. Based on Farthest Point Sampling algorithm, we propose a sampling scheme that theoretically encourages better generalization performance, and results in fast convergence for SGD-based optimization algorithms. Secondly, based on the reflective symmetry of an object, we propose a feature fusion method that alleviates issues due to self-occlusions which makes it difficult to utilize local image features. Our proposed system Ladybird is able to create high quality 3D object reconstructions from a single input image. We evaluate Ladybird on a large scale 3D dataset (ShapeNet) demonstrating highly competitive results in terms of Chamfer distance, Earth Mover's distance and Intersection Over Union (IoU).

Xinhui Song, Tianyang Shi, Tianjia Shao et al.

The automatic intensity estimation of facial action units (AUs) from a single image plays a vital role in facial analysis systems. One big challenge for data-driven AU intensity estimation is the lack of sufficient AU label data. Due to the fact that AU annotation requires strong domain expertise, it is expensive to construct an extensive database to learn deep models. The limited number of labeled AUs as well as identity differences and pose variations further increases the estimation difficulties. Considering all these difficulties, we propose an unsupervised framework GE-Net for facial AU intensity estimation from a single image, without requiring any annotated AU data. Our framework performs differentiable optimization, which iteratively updates the facial parameters (i.e., head pose, AU parameters and identity parameters) to match the input image. GE-Net consists of two modules: a generator and a feature extractor. The generator learns to "render" a face image from a set of facial parameters in a differentiable way, and the feature extractor extracts deep features for measuring the similarity of the rendered image and input real image. After the two modules are trained and fixed, the framework searches optimal facial parameters by minimizing the differences of the extracted features between the rendered image and the input image. Experimental results demonstrate that our method can achieve state-of-the-art results compared with existing methods.

Jiangke Lin, Yi Yuan, Tianjia Shao et al.

3D Morphable Model (3DMM) based methods have achieved great success in recovering 3D face shapes from single-view images. However, the facial textures recovered by such methods lack the fidelity as exhibited in the input images. Recent work demonstrates high-quality facial texture recovering with generative networks trained from a large-scale database of high-resolution UV maps of face textures, which is hard to prepare and not publicly available. In this paper, we introduce a method to reconstruct 3D facial shapes with high-fidelity textures from single-view images in-the-wild, without the need to capture a large-scale face texture database. The main idea is to refine the initial texture generated by a 3DMM based method with facial details from the input image. To this end, we propose to use graph convolutional networks to reconstruct the detailed colors for the mesh vertices instead of reconstructing the UV map. Experiments show that our method can generate high-quality results and outperforms state-of-the-art methods in both qualitative and quantitative comparisons.

Tianyang Shi, Yi Yuan, Changjie Fan et al.

Character customization system is an important component in Role-Playing Games (RPGs), where players are allowed to edit the facial appearance of their in-game characters with their own preferences rather than using default templates. This paper proposes a method for automatically creating in-game characters of players according to an input face photo. We formulate the above "artistic creation" process under a facial similarity measurement and parameter searching paradigm by solving an optimization problem over a large set of physically meaningful facial parameters. To effectively minimize the distance between the created face and the real one, two loss functions, i.e. a "discriminative loss" and a "facial content loss", are specifically designed. As the rendering process of a game engine is not differentiable, a generative network is further introduced as an "imitator" to imitate the physical behavior of the game engine so that the proposed method can be implemented under a neural style transfer framework and the parameters can be optimized by gradient descent. Experimental results demonstrate that our method achieves a high degree of generation similarity between the input face photo and the created in-game character in terms of both global appearance and local details. Our method has been deployed in a new game last year and has now been used by players over 1 million times.

Guanzhong Tian, Yi Yuan, Yong liu

We propose an end to end deep learning approach for generating real-time facial animation from just audio. Specifically, our deep architecture employs deep bidirectional long short-term memory network and attention mechanism to discover the latent representations of time-varying contextual information within the speech and recognize the significance of different information contributed to certain face status. Therefore, our model is able to drive different levels of facial movements at inference and automatically keep up with the corresponding pitch and latent speaking style in the input audio, with no assumption or further human intervention. Evaluation results show that our method could not only generate accurate lip movements from audio, but also successfully regress the speaker's time-varying facial movements.