Mengping Yang

Mengping Yang, Zhe Wang, Ziqiu Chi et al.

Existing few-shot image generation approaches typically employ fusion-based strategies, either on the image or the feature level, to produce new images. However, previous approaches struggle to synthesize high-frequency signals with fine details, deteriorating the synthesis quality. To address this, we propose WaveGAN, a frequency-aware model for few-shot image generation. Concretely, we disentangle encoded features into multiple frequency components and perform low-frequency skip connections to preserve outline and structural information. Then we alleviate the generator's struggles of synthesizing fine details by employing high-frequency skip connections, thus providing informative frequency information to the generator. Moreover, we utilize a frequency L1-loss on the generated and real images to further impede frequency information loss. Extensive experiments demonstrate the effectiveness and advancement of our method on three datasets. Noticeably, we achieve new state-of-the-art with FID 42.17, LPIPS 0.3868, FID 30.35, LPIPS 0.5076, and FID 4.96, LPIPS 0.3822 respectively on Flower, Animal Faces, and VGGFace. GitHub: https://github.com/kobeshegu/ECCV2022_WaveGAN

Mengping Yang, Ceyuan Yang, Yichi Zhang et al.

A good metric, which promises a reliable comparison between solutions, is essential for any well-defined task. Unlike most vision tasks that have per-sample ground-truth, image synthesis tasks target generating unseen data and hence are usually evaluated through a distributional distance between one set of real samples and another set of generated samples. This study presents an empirical investigation into the evaluation of synthesis performance, with generative adversarial networks (GANs) as a representative of generative models. In particular, we make in-depth analyses of various factors, including how to represent a data point in the representation space, how to calculate a fair distance using selected samples, and how many instances to use from each set. Extensive experiments conducted on multiple datasets and settings reveal several important findings. Firstly, a group of models that include both CNN-based and ViT-based architectures serve as reliable and robust feature extractors for measurement evaluation. Secondly, Centered Kernel Alignment (CKA) provides a better comparison across various extractors and hierarchical layers in one model. Finally, CKA is more sample-efficient and enjoys better agreement with human judgment in characterizing the similarity between two internal data correlations. These findings contribute to the development of a new measurement system, which enables a consistent and reliable re-evaluation of current state-of-the-art generative models.

Mengping Yang, Zhe Wang, Ziqiu Chi et al.

Training GANs under limited data often leads to discriminator overfitting and memorization issues, causing divergent training. Existing approaches mitigate the overfitting by employing data augmentations, model regularization, or attention mechanisms. However, they ignore the frequency bias of GANs and take poor consideration towards frequency information, especially high-frequency signals that contain rich details. To fully utilize the frequency information of limited data, this paper proposes FreGAN, which raises the model's frequency awareness and draws more attention to producing high-frequency signals, facilitating high-quality generation. In addition to exploiting both real and generated images' frequency information, we also involve the frequency signals of real images as a self-supervised constraint, which alleviates the GAN disequilibrium and encourages the generator to synthesize adequate rather than arbitrary frequency signals. Extensive results demonstrate the superiority and effectiveness of our FreGAN in ameliorating generation quality in the low-data regime (especially when training data is less than 100). Besides, FreGAN can be seamlessly applied to existing regularization and attention mechanism models to further boost the performance.

Mengping Yang, Zhe Wang

Deep generative models, which target reproducing the given data distribution to produce novel samples, have made unprecedented advancements in recent years. Their technical breakthroughs have enabled unparalleled quality in the synthesis of visual content. However, one critical prerequisite for their tremendous success is the availability of a sufficient number of training samples, which requires massive computation resources. When trained on limited data, generative models tend to suffer from severe performance deterioration due to overfitting and memorization. Accordingly, researchers have devoted considerable attention to develop novel models that are capable of generating plausible and diverse images from limited training data recently. Despite numerous efforts to enhance training stability and synthesis quality in the limited data scenarios, there is a lack of a systematic survey that provides 1) a clear problem definition, critical challenges, and taxonomy of various tasks; 2) an in-depth analysis on the pros, cons, and remain limitations of existing literature; as well as 3) a thorough discussion on the potential applications and future directions in the field of image synthesis under limited data. In order to fill this gap and provide a informative introduction to researchers who are new to this topic, this survey offers a comprehensive review and a novel taxonomy on the development of image synthesis under limited data. In particular, it covers the problem definition, requirements, main solutions, popular benchmarks, and remain challenges in a comprehensive and all-around manner.

Mengping Yang, Zhe Wang, Wenyi Feng et al.

Few-shot image generation, which aims to produce plausible and diverse images for one category given a few images from this category, has drawn extensive attention. Existing approaches either globally interpolate different images or fuse local representations with pre-defined coefficients. However, such an intuitive combination of images/features only exploits the most relevant information for generation, leading to poor diversity and coarse-grained semantic fusion. To remedy this, this paper proposes a novel textural modulation (TexMod) mechanism to inject external semantic signals into internal local representations. Parameterized by the feedback from the discriminator, our TexMod enables more fined-grained semantic injection while maintaining the synthesis fidelity. Moreover, a global structural discriminator (StructD) is developed to explicitly guide the model to generate images with reasonable layout and outline. Furthermore, the frequency awareness of the model is reinforced by encouraging the model to distinguish frequency signals. Together with these techniques, we build a novel and effective model for few-shot image generation. The effectiveness of our model is identified by extensive experiments on three popular datasets and various settings. Besides achieving state-of-the-art synthesis performance on these datasets, our proposed techniques could be seamlessly integrated into existing models for a further performance boost.

Hao Yang, Zhiyu Tan, Jia Gong et al.

We present Omni-Video 2, a scalable and computationally efficient model that connects pretrained multimodal large-language models (MLLMs) with video diffusion models for unified video generation and editing. Our key idea is to exploit the understanding and reasoning capabilities of MLLMs to produce explicit target captions to interpret user instructions. In this way, the rich contextual representations from the understanding model are directly used to guide the generative process, thereby improving performance on complex and compositional editing. Moreover, a lightweight adapter is developed to inject multimodal conditional tokens into pretrained text-to-video diffusion models, allowing maximum reuse of their powerful generative priors in a parameter-efficient manner. Benefiting from these designs, we scale up Omni-Video 2 to a 14B video diffusion model on meticulously curated training data with quality, supporting high quality text-to-video generation and various video editing tasks such as object removal, addition, background change, complex motion editing, \emph{etc.} We evaluate the performance of Omni-Video 2 on the FiVE benchmark for fine-grained video editing and the VBench benchmark for text-to-video generation. The results demonstrate its superior ability to follow complex compositional instructions in video editing, while also achieving competitive or superior quality in video generation tasks.

Mengping Yang, Zhiyu Tan, Binglei Li et al.

Recent breakthroughs in Diffusion Transformers (DiTs) have revolutionized the field of visual synthesis due to their superior scalability. To facilitate DiTs' capability of capturing meaningful internal representations, recent works such as REPA incorporate external pretrained encoders for representation alignment. However, the underlying mechanisms governing representation learning within DiTs are not well understood. To this end, we first systematically investigate the representation dynamics of DiTs. Through analyzing the evolution and influence of internal representations under various settings, we reveal that representation diversity across blocks is a crucial factor for effective learning. Based on this key insight, we propose DiverseDiT, a novel framework that explicitly promotes representation diversity. DiverseDiT incorporates long residual connections to diversify input representations across blocks and a representation diversity loss to encourage blocks to learn distinct features. Extensive experiments on ImageNet 256x256 and 512x512 demonstrate that our DiverseDiT yields consistent performance gains and convergence acceleration when applied to different backbones with various sizes, even when tested on the challenging one-step generation setting. Furthermore, we show that DiverseDiT is complementary to existing representation learning techniques, leading to further performance gains. Our work provides valuable insights into the representation learning dynamics of DiTs and offers a practical approach for enhancing their performance.

Binglei Li, Mengping Yang, Zhiyu Tan et al.

Recent breakthroughs of transformer-based diffusion models, particularly with Multimodal Diffusion Transformers (MMDiT) driven models like FLUX and Qwen Image, have facilitated thrilling experiences in text-to-image generation and editing. To understand the internal mechanism of MMDiT-based models, existing methods tried to analyze the effect of specific components like positional encoding and attention layers. Yet, a comprehensive understanding of how different blocks and their interactions with textual conditions contribute to the synthesis process remains elusive. In this paper, we first develop a systematic pipeline to comprehensively investigate each block's functionality by removing, disabling and enhancing textual hidden-states at corresponding blocks. Our analysis reveals that 1) semantic information appears in earlier blocks and finer details are rendered in later blocks, 2) removing specific blocks is usually less disruptive than disabling text conditions, and 3) enhancing textual conditions in selective blocks improves semantic attributes. Building on these observations, we further propose novel training-free strategies for improved text alignment, precise editing, and acceleration. Extensive experiments demonstrated that our method outperforms various baselines and remains flexible across text-to-image generation, image editing, and inference acceleration. Our method improves T2I-Combench++ from 56.92% to 63.00% and GenEval from 66.42% to 71.63% on SD3.5, without sacrificing synthesis quality. These results advance understanding of MMDiT models and provide valuable insights to unlock new possibilities for further improvements.

Yanbing Zhang, Zhe Wang, Qin Zhou et al.

In light of recent breakthroughs in text-to-image (T2I) generation, particularly with diffusion transformers (DiT), subject-driven technologies are increasingly being employed for high-fidelity customized production that preserves subject identity from reference inputs, enabling thrilling design workflows and engaging entertainment. Existing alternatives typically require either per-subject optimization via trainable text embeddings or training specialized encoders for subject feature extraction on large-scale datasets. Such dependencies on training procedures fundamentally constrain their practical applications. More importantly, current methodologies fail to fully leverage the inherent zero-shot potential of modern diffusion transformers (e.g., the Flux series) for authentic subject-driven synthesis. To bridge this gap, we propose FreeCus, a genuinely training-free framework that activates DiT's capabilities through three key innovations: 1) We introduce a pivotal attention sharing mechanism that captures the subject's layout integrity while preserving crucial editing flexibility. 2) Through a straightforward analysis of DiT's dynamic shifting, we propose an upgraded variant that significantly improves fine-grained feature extraction. 3) We further integrate advanced Multimodal Large Language Models (MLLMs) to enrich cross-modal semantic representations. Extensive experiments reflect that our method successfully unlocks DiT's zero-shot ability for consistent subject synthesis across diverse contexts, achieving state-of-the-art or comparable results compared to approaches that require additional training. Notably, our framework demonstrates seamless compatibility with existing inpainting pipelines and control modules, facilitating more compelling experiences. Our code is available at: https://github.com/Monalissaa/FreeCus.

Yanbing Zhang, Mengping Yang, Qin Zhou et al.

Recent thrilling progress in large-scale text-to-image (T2I) models has unlocked unprecedented synthesis quality of AI-generated content (AIGC) including image generation, 3D and video composition. Further, personalized techniques enable appealing customized production of a novel concept given only several images as reference. However, an intriguing problem persists: Is it possible to capture multiple, novel concepts from one single reference image? In this paper, we identify that existing approaches fail to preserve visual consistency with the reference image and eliminate cross-influence from concepts. To alleviate this, we propose an attention calibration mechanism to improve the concept-level understanding of the T2I model. Specifically, we first introduce new learnable modifiers bound with classes to capture attributes of multiple concepts. Then, the classes are separated and strengthened following the activation of the cross-attention operation, ensuring comprehensive and self-contained concepts. Additionally, we suppress the attention activation of different classes to mitigate mutual influence among concepts. Together, our proposed method, dubbed DisenDiff, can learn disentangled multiple concepts from one single image and produce novel customized images with learned concepts. We demonstrate that our method outperforms the current state of the art in both qualitative and quantitative evaluations. More importantly, our proposed techniques are compatible with LoRA and inpainting pipelines, enabling more interactive experiences.

Zhiyu Tan, Mengping Yang, Luozheng Qin et al.

One critical prerequisite for faithful text-to-image generation is the accurate understanding of text inputs. Existing methods leverage the text encoder of the CLIP model to represent input prompts. However, the pre-trained CLIP model can merely encode English with a maximum token length of 77. Moreover, the model capacity of the text encoder from CLIP is relatively limited compared to Large Language Models (LLMs), which offer multilingual input, accommodate longer context, and achieve superior text representation. In this paper, we investigate LLMs as the text encoder to improve the language understanding in text-to-image generation. Unfortunately, training text-to-image generative model with LLMs from scratch demands significant computational resources and data. To this end, we introduce a three-stage training pipeline that effectively and efficiently integrates the existing text-to-image model with LLMs. Specifically, we propose a lightweight adapter that enables fast training of the text-to-image model using the textual representations from LLMs. Extensive experiments demonstrate that our model supports not only multilingual but also longer input context with superior image generation quality.

Luozheng Qin, Jia Gong, Yuqing Sun et al.

Chain-of-Thought (CoT) reasoning has been widely adopted to enhance Large Language Models (LLMs) by decomposing complex tasks into simpler, sequential subtasks. However, extending CoT to vision-language reasoning tasks remains challenging, as it often requires interpreting transitions of visual states to support reasoning. Existing methods often struggle with this due to limited capacity of modeling visual state transitions or incoherent visual trajectories caused by fragmented architectures. To overcome these limitations, we propose Uni-CoT, a Unified Chain-of-Thought framework that enables coherent and grounded multimodal reasoning within a single unified model. The key idea is to leverage a model capable of both image understanding and generation to reason over visual content and model evolving visual states. However, empowering a unified model to achieve that is non-trivial, given the high computational cost and the burden of training. To address this, Uni-CoT introduces a novel two-level reasoning paradigm: A Macro-Level CoT for high-level task planning and A Micro-Level CoT for subtask execution. This design significantly reduces the computational overhead. Furthermore, we introduce a structured training paradigm that combines interleaved image-text supervision for macro-level CoT with multi-task objectives for micro-level CoT. Together, these innovations allow Uni-CoT to perform scalable and coherent multi-modal reasoning. Furthermore, thanks to our design, all experiments can be efficiently completed using only 8 A100 GPUs with 80GB VRAM each. Experimental results on reasoning-driven image generation benchmark (WISE) and editing benchmarks (RISE and KRIS) indicates that Uni-CoT demonstrates SOTA performance and strong generalization, establishing Uni-CoT as a promising solution for multi-modal reasoning. Project Page and Code: https://sais-fuxi.github.io/projects/uni-cot/

Zhiyu Tan, Hao Yang, Luozheng Qin et al.

Notable breakthroughs in unified understanding and generation modeling have led to remarkable advancements in image understanding, reasoning, production and editing, yet current foundational models predominantly focus on processing images, creating a gap in the development of unified models for video understanding and generation. This report presents Omni-Video, an efficient and effective unified framework for video understanding, generation, as well as instruction-based editing. Our key insight is to teach existing multimodal large language models (MLLMs) to produce continuous visual clues that are used as the input of diffusion decoders, which produce high-quality videos conditioned on these visual clues. To fully unlock the potential of our system for unified video modeling, we integrate several technical improvements: 1) a lightweight architectural design that respectively attaches a vision head on the top of MLLMs and a adapter before the input of diffusion decoders, the former produce visual tokens for the latter, which adapts these visual tokens to the conditional space of diffusion decoders; and 2) an efficient multi-stage training scheme that facilitates a fast connection between MLLMs and diffusion decoders with limited data and computational resources. We empirically demonstrate that our model exhibits satisfactory generalization abilities across video generation, editing and understanding tasks.

Luozheng Qin, Zhiyu Tan, Mengping Yang et al.

Video Detailed Captioning (VDC) is a crucial task for vision-language bridging, enabling fine-grained descriptions of complex video content. In this paper, we first comprehensively benchmark current state-of-the-art approaches and systematically identified two critical limitations: biased capability towards specific captioning aspect and misalignment with human preferences. To address these deficiencies, we propose Cockatiel, a novel three-stage training pipeline that ensembles synthetic and human-aligned training for improving VDC performance. In the first stage, we derive a scorer from a meticulously annotated dataset to select synthetic captions high-performing on certain fine-grained video-caption alignment and human-preferred while disregarding others. Then, we train Cockatiel-13B, using this curated dataset to infuse it with assembled model strengths and human preferences. Finally, we further distill Cockatiel-8B from Cockatiel-13B for the ease of usage. Extensive quantitative and qualitative experiments reflect the effectiveness of our method, as we not only set new state-of-the-art performance on VDCSCORE in a dimension-balanced way but also surpass leading alternatives on human preference by a large margin as depicted by the human evaluation results.

Xiaomeng Yang, Zhiyu Tan, Xiaohui Zhong et al.

Scientific discovery increasingly relies on integrating heterogeneous, high-dimensional data across disciplines nowadays. While AI models have achieved notable success across various scientific domains, they typically remain domain-specific or lack the capability of simultaneously understanding and generating multimodal scientific data, particularly for high-dimensional data. Yet, many pressing global challenges and scientific problems are inherently cross-disciplinary and require coordinated progress across multiple fields. Here, we present FuXi-Uni, a native unified multimodal model for scientific understanding and high-fidelity generation across scientific domains within a single architecture. Specifically, FuXi-Uni aligns cross-disciplinary scientific tokens within natural language tokens and employs science decoder to reconstruct scientific tokens, thereby supporting both natural language conversation and scientific numerical prediction. Empirically, we validate FuXi-Uni in Earth science and Biomedicine. In Earth system modeling, the model supports global weather forecasting, tropical cyclone (TC) forecast editing, and spatial downscaling driven by only language instructions. FuXi-Uni generates 10-day global forecasts at 0.25° resolution that outperform the SOTA physical forecasting system. It shows superior performance for both TC track and intensity prediction relative to the SOTA physical model, and generates high-resolution regional weather fields that surpass standard interpolation baselines. Regarding biomedicine, FuXi-Uni outperforms leading multimodal large language models on multiple biomedical visual question answering benchmarks. By unifying heterogeneous scientific modalities within a native shared latent space while maintaining strong domain-specific performance, FuXi-Uni provides a step forward more general-purpose, multimodal scientific models.

Mengping Yang, Zhe Wang, Ziqiu Chi et al.

Generative adversarial networks (GANs) have made remarkable achievements in synthesizing images in recent years. Typically, training GANs requires massive data, and the performance of GANs deteriorates significantly when training data is limited. To improve the synthesis performance of GANs in low-data regimes, existing approaches use various data augmentation techniques to enlarge the training sets. However, it is identified that these augmentation techniques may leak or even alter the data distribution. To remedy this, we propose an adversarial semantic augmentation (ASA) technique to enlarge the training data at the semantic level instead of the image level. Concretely, considering semantic features usually encode informative information of images, we estimate the covariance matrices of semantic features for both real and generated images to find meaningful transformation directions. Such directions translate original features to another semantic representation, e.g., changing the backgrounds or expressions of the human face dataset. Moreover, we derive an upper bound of the expected adversarial loss. By optimizing the upper bound, our semantic augmentation is implicitly achieved. Such design avoids redundant sampling of the augmented features and introduces negligible computation overhead, making our approach computation efficient. Extensive experiments on both few-shot and large-scale datasets demonstrate that our method consistently improve the synthesis quality under various data regimes, and further visualized and analytic results suggesting satisfactory versatility of our proposed method.

Zhiyu Tan, WenXu Qian, Hesen Chen et al.

Diffusion models have established themselves as the de facto primary paradigm in visual generative modeling, revolutionizing the field through remarkable success across various diverse applications ranging from high-quality image synthesis to temporal aware video generation. Despite these advancements, three fundamental limitations persist, including 1) discrepancy between training and inference processes, 2) progressive information leakage throughout the noise corruption procedures, and 3) inherent constraints preventing effective integration of modern optimization criteria like perceptual and adversarial loss. To mitigate these critical challenges, we in this paper present a novel end-to-end learning paradigm that establishes direct optimization from the final generated samples to initial noises. Our proposed End-to-End Differentiable Diffusion, dubbed E2ED^2, introduces several key improvements: it eliminates the sequential training-sampling mismatch and intermediate information leakage via conceptualizing training as a direct transformation from isotropic Gaussian noise to the target data distribution. Additionally, such training framework enables seamless incorporation of adversarial and perceptual losses into the core optimization objective. Comprehensive evaluation across standard benchmarks including COCO30K and HW30K reveals that our method achieves substantial performance gains in terms of Fréchet Inception Distance (FID) and CLIP score, even with fewer sampling steps (less than 4). Our findings highlight that the end-to-end mechanism might pave the way for more robust and efficient solutions, \emph{i.e.,} combining diffusion stability with GAN-like discriminative optimization in an end-to-end manner.

Zhiyu Tan, Xiaomeng Yang, Luozheng Qin et al.

The recent advancements in text-to-image generative models have been remarkable. Yet, the field suffers from a lack of evaluation metrics that accurately reflect the performance of these models, particularly lacking fine-grained metrics that can guide the optimization of the models. In this paper, we propose EvalAlign, a metric characterized by its accuracy, stability, and fine granularity. Our approach leverages the capabilities of Multimodal Large Language Models (MLLMs) pre-trained on extensive data. We develop evaluation protocols that focus on two key dimensions: image faithfulness and text-image alignment. Each protocol comprises a set of detailed, fine-grained instructions linked to specific scoring options, enabling precise manual scoring of the generated images. We supervised fine-tune (SFT) the MLLM to align with human evaluative judgments, resulting in a robust evaluation model. Our evaluation across 24 text-to-image generation models demonstrate that EvalAlign not only provides superior metric stability but also aligns more closely with human preferences than existing metrics, confirming its effectiveness and utility in model assessment.