Fan Tang

Xiaoyue Mi, Fan Tang, Zonghan Yang et al. · tsinghua

Despite the remarkable advances that have been made in continual learning, the adversarial vulnerability of such methods has not been fully discussed. We delve into the adversarial robustness of memory-based continual learning algorithms and observe limited robustness improvement by directly applying adversarial training techniques. Preliminary studies reveal the twin challenges for building adversarial robust continual learners: accelerated forgetting in continual learning and gradient obfuscation in adversarial robustness. In this study, we put forward a novel adversarial robust memory-based continual learner that adjusts data logits to mitigate the forgetting of pasts caused by adversarial samples. Furthermore, we devise a gradient-based data selection mechanism to overcome the gradient obfuscation caused by limited stored data. The proposed approach can widely integrate with existing memory-based continual learning as well as adversarial training algorithms in a plug-and-play way. Extensive experiments on Split-CIFAR10/100 and Split-Tiny-ImageNet demonstrate the effectiveness of our approach, achieving up to 8.13% higher accuracy for adversarial data.

Yingying Deng, Xiangyu He, Fan Tang et al.

Despite the remarkable progress in image style transfer, formulating style in the context of art is inherently subjective and challenging. In contrast to existing learning/tuning methods, this study shows that vanilla diffusion models can directly extract style information and seamlessly integrate the generative prior into the content image without retraining. Specifically, we adopt dual denoising paths to represent content/style references in latent space and then guide the content image denoising process with style latent codes. We further reveal that the cross-attention mechanism in latent diffusion models tends to blend the content and style images, resulting in stylized outputs that deviate from the original content image. To overcome this limitation, we introduce a cross-attention rearrangement strategy. Through theoretical analysis and experiments, we demonstrate the effectiveness and superiority of the diffusion-based $\underline{Z}$ero-shot $\underline{S}$tyle $\underline{T}$ransfer via $\underline{A}$ttention $\underline{R}$earrangement, Z-STAR.

Ziyao Huang, Fan Tang, Yong Zhang et al.

Despite the remarkable process of talking-head-based avatar-creating solutions, directly generating anchor-style videos with full-body motions remains challenging. In this study, we propose Make-Your-Anchor, a novel system necessitating only a one-minute video clip of an individual for training, subsequently enabling the automatic generation of anchor-style videos with precise torso and hand movements. Specifically, we finetune a proposed structure-guided diffusion model on input video to render 3D mesh conditions into human appearances. We adopt a two-stage training strategy for the diffusion model, effectively binding movements with specific appearances. To produce arbitrary long temporal video, we extend the 2D U-Net in the frame-wise diffusion model to a 3D style without additional training cost, and a simple yet effective batch-overlapped temporal denoising module is proposed to bypass the constraints on video length during inference. Finally, a novel identity-specific face enhancement module is introduced to improve the visual quality of facial regions in the output videos. Comparative experiments demonstrate the effectiveness and superiority of the system in terms of visual quality, temporal coherence, and identity preservation, outperforming SOTA diffusion/non-diffusion methods. Project page: \url{https://github.com/ICTMCG/Make-Your-Anchor}.

Yingying Deng, Xiangyu He, Changwang Mei et al.

Though Rectified Flows (ReFlows) with distillation offers a promising way for fast sampling, its fast inversion transforms images back to structured noise for recovery and following editing remains unsolved. This paper introduces FireFlow, a simple yet effective zero-shot approach that inherits the startling capacity of ReFlow-based models (such as FLUX) in generation while extending its capabilities to accurate inversion and editing in $8$ steps. We first demonstrate that a carefully designed numerical solver is pivotal for ReFlow inversion, enabling accurate inversion and reconstruction with the precision of a second-order solver while maintaining the practical efficiency of a first-order Euler method. This solver achieves a $3\times$ runtime speedup compared to state-of-the-art ReFlow inversion and editing techniques, while delivering smaller reconstruction errors and superior editing results in a training-free mode. The code is available at $\href{https://github.com/HolmesShuan/FireFlow}{this URL}$.

Ziyi Xu, Ziyao Huang, Juan Cao et al.

The generation of anchor-style product promotion videos presents promising opportunities in e-commerce, advertising, and consumer engagement. Despite advancements in pose-guided human video generation, creating product promotion videos remains challenging. In addressing this challenge, we identify the integration of human-object interactions (HOI) into pose-guided human video generation as a core issue. To this end, we introduce AnchorCrafter, a novel diffusion-based system designed to generate 2D videos featuring a target human and a customized object, achieving high visual fidelity and controllable interactions. Specifically, we propose two key innovations: the HOI-appearance perception, which enhances object appearance recognition from arbitrary multi-view perspectives and disentangles object and human appearance, and the HOI-motion injection, which enables complex human-object interactions by overcoming challenges in object trajectory conditioning and inter-occlusion management. Extensive experiments show that our system improves object appearance preservation by 7.5\% and doubles the object localization accuracy compared to existing state-of-the-art approaches. It also outperforms existing approaches in maintaining human motion consistency and high-quality video generation. Project page including data, code, and Huggingface demo: https://github.com/cangcz/AnchorCrafter.

Haipeng Fang, Zhihao Sun, Ziyao Huang et al.

The advancement of generative AI has extended to the realm of Human Dance Generation, demonstrating superior generative capacities. However, current methods still exhibit deficiencies in achieving spatiotemporal consistency, resulting in artifacts like ghosting, flickering, and incoherent motions. In this paper, we present Dance-Your-Latents, a framework that makes latents dance coherently following motion flow to generate consistent dance videos. Firstly, considering that each constituent element moves within a confined space, we introduce spatial-temporal subspace-attention blocks that decompose the global space into a combination of regular subspaces and efficiently model the spatiotemporal consistency within these subspaces. This module enables each patch pay attention to adjacent areas, mitigating the excessive dispersion of long-range attention. Furthermore, observing that body part's movement is guided by pose control, we design motion flow guided subspace align & restore. This method enables the attention to be computed on the irregular subspace along the motion flow. Experimental results in TikTok dataset demonstrate that our approach significantly enhances spatiotemporal consistency of the generated videos.

Yiguo Jiang, Xiaodong Cun, Yong Zhang et al.

Emotional talking head synthesis aims to generate talking portrait videos with vivid expressions. Existing methods still exhibit limitations in control flexibility, motion naturalness, and expression quality. Moreover, currently available datasets are primarily collected in lab settings, further exacerbating these shortcomings. Consequently, these limitations substantially hinder practical applications in real-world scenarios. To address these challenges, we propose EmoCAST, a diffusion-based framework with two key modules for precise text-driven emotional synthesis. In appearance modeling, emotional prompts are integrated through a text-guided decoupled emotive module, enhancing the spatial knowledge to improve emotion comprehension. To improve the relationship between audio and emotion, we introduce an emotive audio attention module to capture the interplay between controlled emotion and driving audio, generating emotion-aware features to guide more precise facial motion synthesis. Additionally, we construct an emotional talking head dataset with comprehensive emotive text descriptions to optimize the framework's performance. Based on the proposed dataset, we propose an emotion-aware sampling training strategy and a progressive functional training strategy that further improve the model's ability to capture nuanced expressive features and achieve accurate lip-synchronization. Overall, EmoCAST achieves state-of-the-art performance in generating realistic, emotionally expressive, and audio-synchronized talking-head videos. Project Page: https://github.com/GVCLab/EmoCAST

Nisha Huang, Weiming Dong, Yuxin Zhang et al.

Although remarkable progress has been made in image style transfer, style is just one of the components of artistic paintings. Directly transferring extracted style features to natural images often results in outputs with obvious synthetic traces. This is because key painting attributes including layout, perspective, shape, and semantics often cannot be conveyed and expressed through style transfer. Large-scale pretrained text-to-image generation models have demonstrated their capability to synthesize a vast amount of high-quality images. However, even with extensive textual descriptions, it is challenging to fully express the unique visual properties and details of paintings. Moreover, generic models often disrupt the overall artistic effect when modifying specific areas, making it more complicated to achieve a unified aesthetic in artworks. Our main novel idea is to integrate multimodal semantic information as a synthesis guide into artworks, rather than transferring style to the real world. We also aim to reduce the disruption to the harmony of artworks while simplifying the guidance conditions. Specifically, we propose an innovative multi-task unified framework called CreativeSynth, based on the diffusion model with the ability to coordinate multimodal inputs. CreativeSynth combines multimodal features with customized attention mechanisms to seamlessly integrate real-world semantic content into the art domain through Cross-Art-Attention for aesthetic maintenance and semantic fusion. We demonstrate the results of our method across a wide range of different art categories, proving that CreativeSynth bridges the gap between generative models and artistic expression. Code and results are available at https://github.com/haha-lisa/CreativeSynth.

Xingjia Pan, Yingguo Gao, Zhiwen Lin et al.

Weakly supervised object localization(WSOL) remains an open problem given the deficiency of finding object extent information using a classification network. Although prior works struggled to localize objects through various spatial regularization strategies, we argue that how to extract object structural information from the trained classification network is neglected. In this paper, we propose a two-stage approach, termed structure-preserving activation (SPA), toward fully leveraging the structure information incorporated in convolutional features for WSOL. First, a restricted activation module (RAM) is designed to alleviate the structure-missing issue caused by the classification network on the basis of the observation that the unbounded classification map and global average pooling layer drive the network to focus only on object parts. Second, we designed a post-process approach, termed self-correlation map generating (SCG) module to obtain structure-preserving localization maps on the basis of the activation maps acquired from the first stage. Specifically, we utilize the high-order self-correlation (HSC) to extract the inherent structural information retained in the learned model and then aggregate HSC of multiple points for precise object localization. Extensive experiments on two publicly available benchmarks including CUB-200-2011 and ILSVRC show that the proposed SPA achieves substantial and consistent performance gains compared with baseline approaches.Code and models are available at https://github.com/Panxjia/SPA_CVPR2021

Yu Xu, Fan Tang, Juan Cao et al.

Diffusion Transformers (DiTs) have exhibited robust capabilities in image generation tasks. However, accurate text-guided image editing for multimodal DiTs (MM-DiTs) still poses a significant challenge. Unlike UNet-based structures that could utilize self/cross-attention maps for semantic editing, MM-DiTs inherently lack support for explicit and consistent incorporated text guidance, resulting in semantic misalignment between the edited results and texts. In this study, we disclose the sensitivity of different attention heads to different image semantics within MM-DiTs and introduce HeadRouter, a training-free image editing framework that edits the source image by adaptively routing the text guidance to different attention heads in MM-DiTs. Furthermore, we present a dual-token refinement module to refine text/image token representations for precise semantic guidance and accurate region expression. Experimental results on multiple benchmarks demonstrate HeadRouter's performance in terms of editing fidelity and image quality.

Yu Xu, Fan Tang, Juan Cao et al.

Personalized generation paradigms empower designers to customize visual intellectual properties with the help of textual descriptions by tuning or adapting pre-trained text-to-image models on a few images. Recent works explore approaches for concurrently customizing both content and detailed visual style appearance. However, these existing approaches often generate images where the content and style are entangled. In this study, we reconsider the customization of content and style concepts from the perspective of parameter space construction. Unlike existing methods that utilize a shared parameter space for content and style, we propose a learning framework that separates the parameter space to facilitate individual learning of content and style, thereby enabling disentangled content and style. To achieve this goal, we introduce "partly learnable projection" (PLP) matrices to separate the original adapters into divided sub-parameter spaces. We propose "break-for-make" customization learning pipeline based on PLP, which is simple yet effective. We break the original adapters into "up projection" and "down projection", train content and style PLPs individually with the guidance of corresponding textual prompts in the separate adapters, and maintain generalization by employing a multi-correspondence projection learning strategy. Based on the adapters broken apart for separate training content and style, we then make the entity parameter space by reconstructing the content and style PLPs matrices, followed by fine-tuning the combined adapter to generate the target object with the desired appearance. Experiments on various styles, including textures, materials, and artistic style, show that our method outperforms state-of-the-art single/multiple concept learning pipelines in terms of content-style-prompt alignment.

You Wu, Kean Liu, Xiaoyue Mi et al.

Concept personalization methods enable large text-to-image models to learn specific subjects (e.g., objects/poses/3D models) and synthesize renditions in new contexts. Given that the image references are highly biased towards visual attributes, state-of-the-art personalization models tend to overfit the whole subject and cannot disentangle visual characteristics in pixel space. In this study, we proposed a more challenging setting, namely fine-grained visual appearance personalization. Different from existing methods, we allow users to provide a sentence describing the desired attributes. A novel decoupled self-augmentation strategy is proposed to generate target-related and non-target samples to learn user-specified visual attributes. These augmented data allow for refining the model's understanding of the target attribute while mitigating the impact of unrelated attributes. At the inference stage, adjustments are conducted on semantic space through the learned target and non-target embeddings to further enhance the disentanglement of target attributes. Extensive experiments on various kinds of visual attributes with SOTA personalization methods show the ability of the proposed method to mimic target visual appearance in novel contexts, thus improving the controllability and flexibility of personalization.

Ziyao Huang, Zixiang Zhou, Juan Cao et al.

To address key limitations in human-object interaction (HOI) video generation -- specifically the reliance on curated motion data, limited generalization to novel objects/scenarios, and restricted accessibility -- we introduce HunyuanVideo-HOMA, a weakly conditioned multimodal-driven framework. HunyuanVideo-HOMA enhances controllability and reduces dependency on precise inputs through sparse, decoupled motion guidance. It encodes appearance and motion signals into the dual input space of a multimodal diffusion transformer (MMDiT), fusing them within a shared context space to synthesize temporally consistent and physically plausible interactions. To optimize training, we integrate a parameter-space HOI adapter initialized from pretrained MMDiT weights, preserving prior knowledge while enabling efficient adaptation, and a facial cross-attention adapter for anatomically accurate audio-driven lip synchronization. Extensive experiments confirm state-of-the-art performance in interaction naturalness and generalization under weak supervision. Finally, HunyuanVideo-HOMA demonstrates versatility in text-conditioned generation and interactive object manipulation, supported by a user-friendly demo interface. The project page is at https://anonymous.4open.science/w/homa-page-0FBE/.

Zijun Zhou, Yingying Deng, Xiangyu He et al.

Many real-world applications, such as interactive photo retouching, artistic content creation, and product design, require flexible and iterative image editing. However, existing image editing methods primarily focus on achieving the desired modifications in a single step, which often struggles with ambiguous user intent, complex transformations, or the need for progressive refinements. As a result, these methods frequently produce inconsistent outcomes or fail to meet user expectations. To address these challenges, we propose a multi-turn image editing framework that enables users to iteratively refine their edits, progressively achieving more satisfactory results. Our approach leverages flow matching for accurate image inversion and a dual-objective Linear Quadratic Regulators (LQR) for stable sampling, effectively mitigating error accumulation. Additionally, by analyzing the layer-wise roles of transformers, we introduce a adaptive attention highlighting method that enhances editability while preserving multi-turn coherence. Extensive experiments demonstrate that our framework significantly improves edit success rates and visual fidelity compared to existing methods.

Hairui Ren, Fan Tang, He Zhao et al.

Fine-tuning vision-language models (VLMs) with large amounts of unlabeled data has recently garnered significant interest. However, a key challenge remains the lack of high-quality pseudo-labeled data. Current pseudo-labeling strategies often struggle with mismatches between semantic and visual information, leading to sub-optimal performance of unsupervised prompt learning (UPL) methods. In this paper, we introduce a simple yet effective approach called \textbf{A}ugmenting D\textbf{i}scriminative \textbf{R}ichness via Diffusions (AiR), toward learning a richer discriminating way to represent the class comprehensively and thus facilitate classification. Specifically, our approach includes a pseudo-label generation module that leverages high-fidelity synthetic samples to create an auxiliary classifier, which captures richer visual variation, bridging text-image-pair classification to a more robust image-image-pair classification. Additionally, we exploit the diversity of diffusion-based synthetic samples to enhance prompt learning, providing greater information for semantic-visual alignment. Extensive experiments on five public benchmarks, including RESISC45 and Flowers102, and across three learning paradigms-UL, SSL, and TRZSL-demonstrate that AiR achieves substantial and consistent performance improvements over state-of-the-art unsupervised prompt learning methods.

Xiaoyue Mi, Fan Tang, Juan Cao et al.

Visual generation models have achieved remarkable progress in computer graphics applications but still face significant challenges in real-world deployment. Current assessment approaches for visual generation tasks typically follow an isolated three-phase framework: test input collection, model output generation, and user assessment. These fashions suffer from fixed coverage, evolving difficulty, and data leakage risks, limiting their effectiveness in comprehensively evaluating increasingly complex generation models. To address these limitations, we propose DyEval, an LLM-powered dynamic interactive visual assessment framework that facilitates collaborative evaluation between humans and generative models for text-to-image systems. DyEval features an intuitive visual interface that enables users to interactively explore and analyze model behaviors, while adaptively generating hierarchical, fine-grained, and diverse textual inputs to continuously probe the capability boundaries of the models based on their feedback. Additionally, to provide interpretable analysis for users to further improve tested models, we develop a contextual reflection module that mines failure triggers of test inputs and reflects model potential failure patterns supporting in-depth analysis using the logical reasoning ability of LLM. Qualitative and quantitative experiments demonstrate that DyEval can effectively help users identify max up to 2.56 times generation failures than conventional methods, and uncover complex and rare failure patterns, such as issues with pronoun generation and specific cultural context generation. Our framework provides valuable insights for improving generative models and has broad implications for advancing the reliability and capabilities of visual generation systems across various domains.

Shulan Ruan, Yong Zhang, Kun Zhang et al.

Text-to-image synthesis refers to generating an image from a given text description, the key goal of which lies in photo realism and semantic consistency. Previous methods usually generate an initial image with sentence embedding and then refine it with fine-grained word embedding. Despite the significant progress, the 'aspect' information (e.g., red eyes) contained in the text, referring to several words rather than a word that depicts 'a particular part or feature of something', is often ignored, which is highly helpful for synthesizing image details. How to make better utilization of aspect information in text-to-image synthesis still remains an unresolved challenge. To address this problem, in this paper, we propose a Dynamic Aspect-awarE GAN (DAE-GAN) that represents text information comprehensively from multiple granularities, including sentence-level, word-level, and aspect-level. Moreover, inspired by human learning behaviors, we develop a novel Aspect-aware Dynamic Re-drawer (ADR) for image refinement, in which an Attended Global Refinement (AGR) module and an Aspect-aware Local Refinement (ALR) module are alternately employed. AGR utilizes word-level embedding to globally enhance the previously generated image, while ALR dynamically employs aspect-level embedding to refine image details from a local perspective. Finally, a corresponding matching loss function is designed to ensure the text-image semantic consistency at different levels. Extensive experiments on two well-studied and publicly available datasets (i.e., CUB-200 and COCO) demonstrate the superiority and rationality of our method.

Cong Wang, Fan Tang, Yong Zhang et al.

Regional facial image synthesis conditioned on semantic mask has achieved great success using generative adversarial networks. However, the appearance of different regions may be inconsistent with each other when conducting regional image editing. In this paper, we focus on the problem of harmonized regional style transfer and manipulation for facial images. The proposed approach supports regional style transfer and manipulation at the same time. A multi-scale encoder and style mapping networks are proposed in our work. The encoder is responsible for extracting regional styles of real faces. Style mapping networks generate styles from random samples for all facial regions. As the key part of our work, we propose a multi-region style attention module to adapt the multiple regional style embeddings from a reference image to a target image for generating harmonious and plausible results. Furthermore, we propose a new metric "harmony score" and conduct experiments in a challenging setting: three widely used face datasets are involved and we test the model by transferring the regional facial appearance between datasets. Images in different datasets are usually quite different, which makes the inconsistency between target and reference regions more obvious. Results show that our model can generate reliable style transfer and multi-modal manipulation results compared with SOTAs. Furthermore, we show two face editing applications using the proposed approach.

Yingying Deng, Fan Tang, Weiming Dong et al.

Video style transfer is getting more attention in AI community for its numerous applications such as augmented reality and animation productions. Compared with traditional image style transfer, performing this task on video presents new challenges: how to effectively generate satisfactory stylized results for any specified style, and maintain temporal coherence across frames at the same time. Towards this end, we propose Multi-Channel Correction network (MCCNet), which can be trained to fuse the exemplar style features and input content features for efficient style transfer while naturally maintaining the coherence of input videos. Specifically, MCCNet works directly on the feature space of style and content domain where it learns to rearrange and fuse style features based on their similarity with content features. The outputs generated by MCC are features containing the desired style patterns which can further be decoded into images with vivid style textures. Moreover, MCCNet is also designed to explicitly align the features to input which ensures the output maintains the content structures as well as the temporal continuity. To further improve the performance of MCCNet under complex light conditions, we also introduce the illumination loss during training. Qualitative and quantitative evaluations demonstrate that MCCNet performs well in both arbitrary video and image style transfer tasks.

Minxuan Lin, Fan Tang, Weiming Dong et al.

Multimodal and multi-domain stylization are two important problems in the field of image style transfer. Currently, there are few methods that can perform both multimodal and multi-domain stylization simultaneously. In this paper, we propose a unified framework for multimodal and multi-domain style transfer with the support of both exemplar-based reference and randomly sampled guidance. The key component of our method is a novel style distribution alignment module that eliminates the explicit distribution gaps between various style domains and reduces the risk of mode collapse. The multimodal diversity is ensured by either guidance from multiple images or random style code, while the multi-domain controllability is directly achieved by using a domain label. We validate our proposed framework on painting style transfer with a variety of different artistic styles and genres. Qualitative and quantitative comparisons with state-of-the-art methods demonstrate that our method can generate high-quality results of multi-domain styles and multimodal instances with reference style guidance or random sampled style.

Yingying Deng, Fan Tang, Weiming Dong et al.

Arbitrary style transfer is a significant topic with research value and application prospect. A desired style transfer, given a content image and referenced style painting, would render the content image with the color tone and vivid stroke patterns of the style painting while synchronously maintaining the detailed content structure information. Style transfer approaches would initially learn content and style representations of the content and style references and then generate the stylized images guided by these representations. In this paper, we propose the multi-adaptation network which involves two self-adaptation (SA) modules and one co-adaptation (CA) module: the SA modules adaptively disentangle the content and style representations, i.e., content SA module uses position-wise self-attention to enhance content representation and style SA module uses channel-wise self-attention to enhance style representation; the CA module rearranges the distribution of style representation based on content representation distribution by calculating the local similarity between the disentangled content and style features in a non-local fashion. Moreover, a new disentanglement loss function enables our network to extract main style patterns and exact content structures to adapt to various input images, respectively. Various qualitative and quantitative experiments demonstrate that the proposed multi-adaptation network leads to better results than the state-of-the-art style transfer methods.

Minxuan Lin, Yingying Deng, Fan Tang et al.

Controllable painting generation plays a pivotal role in image stylization. Currently, the control way of style transfer is subject to exemplar-based reference or a random one-hot vector guidance. Few works focus on decoupling the intrinsic properties of painting as control conditions, e.g., artist, genre and period. Under this circumstance, we propose a novel framework adopting multiple attributes from the painting to control the stylized results. An asymmetrical cycle structure is equipped to preserve the fidelity, associating with style preserving and attribute regression loss to keep the unique distinction of colors and textures between domains. Several qualitative and quantitative results demonstrate the effect of the combinations of multiple attributes and achieve satisfactory performance.

Fan Tang, Weiming Dong, Yiping Meng et al.

Real-world applications could benefit from the ability to automatically retarget an image to different aspect ratios and resolutions, while preserving its visually and semantically important content. However, not all images can be equally well processed that way. In this work, we introduce the notion of image retargetability to describe how well a particular image can be handled by content-aware image retargeting. We propose to learn a deep convolutional neural network to rank photo retargetability in which the relative ranking of photo retargetability is directly modeled in the loss function. Our model incorporates joint learning of meaningful photographic attributes and image content information which can help regularize the complicated retargetability rating problem. To train and analyze this model, we have collected a database which contains retargetability scores and meaningful image attributes assigned by six expert raters. Experiments demonstrate that our unified model can generate retargetability rankings that are highly consistent with human labels. To further validate our model, we show applications of image retargetability in retargeting method selection, retargeting method assessment and photo collage generation.