Chaoyue Wang

Zuopeng Yang, Daqing Liu, Chaoyue Wang et al.

We present a method that achieves state-of-the-art results on challenging (few-shot) layout-to-image generation tasks by accurately modeling textures, structures and relationships contained in a complex scene. After compressing RGB images into patch tokens, we propose the Transformer with Focal Attention (TwFA) for exploring dependencies of object-to-object, object-to-patch and patch-to-patch. Compared to existing CNN-based and Transformer-based generation models that entangled modeling on pixel-level&patch-level and object-level&patch-level respectively, the proposed focal attention predicts the current patch token by only focusing on its highly-related tokens that specified by the spatial layout, thereby achieving disambiguation during training. Furthermore, the proposed TwFA largely increases the data efficiency during training, therefore we propose the first few-shot complex scene generation strategy based on the well-trained TwFA. Comprehensive experiments show the superiority of our method, which significantly increases both quantitative metrics and qualitative visual realism with respect to state-of-the-art CNN-based and transformer-based methods. Code is available at https://github.com/JohnDreamer/TwFA.

Youjian Zhang, Chaoyue Wang, Dacheng Tao

Real-world dynamic scene deblurring has long been a challenging task since paired blurry-sharp training data is unavailable. Conventional Maximum A Posteriori estimation and deep learning-based deblurring methods are restricted by handcrafted priors and synthetic blurry-sharp training pairs respectively, thereby failing to generalize to real dynamic blurriness. To this end, we propose a Neural Maximum A Posteriori (NeurMAP) estimation framework for training neural networks to recover blind motion information and sharp content from unpaired data. The proposed NeruMAP consists of a motion estimation network and a deblurring network which are trained jointly to model the (re)blurring process (i.e. likelihood function). Meanwhile, the motion estimation network is trained to explore the motion information in images by applying implicit dynamic motion prior, and in return enforces the deblurring network training (i.e. providing sharp image prior). The proposed NeurMAP is an orthogonal approach to existing deblurring neural networks, and is the first framework that enables training image deblurring networks on unpaired datasets. Experiments demonstrate our superiority on both quantitative metrics and visual quality over state-of-the-art methods. Codes are available on https://github.com/yjzhang96/NeurMAP-deblur.

Zuopeng Yang, Tianshu Chu, Xin Lin et al.

The recent success of text-to-image generation diffusion models has also revolutionized semantic image editing, enabling the manipulation of images based on query/target texts. Despite these advancements, a significant challenge lies in the potential introduction of contextual prior bias in pre-trained models during image editing, e.g., making unexpected modifications to inappropriate regions. To address this issue, we present a novel approach called Dual-Cycle Diffusion, which generates an unbiased mask to guide image editing. The proposed model incorporates a Bias Elimination Cycle that consists of both a forward path and an inverted path, each featuring a Structural Consistency Cycle to ensure the preservation of image content during the editing process. The forward path utilizes the pre-trained model to produce the edited image, while the inverted path converts the result back to the source image. The unbiased mask is generated by comparing differences between the processed source image and the edited image to ensure that both conform to the same distribution. Our experiments demonstrate the effectiveness of the proposed method, as it significantly improves the D-CLIP score from 0.272 to 0.283. The code will be available at https://github.com/JohnDreamer/DualCycleDiffsion.

Jing Zhao, Heliang Zheng, Chaoyue Wang et al.

The advent of open-source AI communities has produced a cornucopia of powerful text-guided diffusion models that are trained on various datasets. While few explorations have been conducted on ensembling such models to combine their strengths. In this work, we propose a simple yet effective method called Saliency-aware Noise Blending (SNB) that can empower the fused text-guided diffusion models to achieve more controllable generation. Specifically, we experimentally find that the responses of classifier-free guidance are highly related to the saliency of generated images. Thus we propose to trust different models in their areas of expertise by blending the predicted noises of two diffusion models in a saliency-aware manner. SNB is training-free and can be completed within a DDIM sampling process. Additionally, it can automatically align the semantics of two noise spaces without requiring additional annotations such as masks. Extensive experiments show the impressive effectiveness of SNB in various applications. Project page is available at https://magicfusion.github.io/.

Wenquan Lu, Yufei Xu, Jing Zhang et al.

Diffusion models have achieved remarkable success in generating realistic images but suffer from generating accurate human hands, such as incorrect finger counts or irregular shapes. This difficulty arises from the complex task of learning the physical structure and pose of hands from training images, which involves extensive deformations and occlusions. For correct hand generation, our paper introduces a lightweight post-processing solution called $\textbf{HandRefiner}$. HandRefiner employs a conditional inpainting approach to rectify malformed hands while leaving other parts of the image untouched. We leverage the hand mesh reconstruction model that consistently adheres to the correct number of fingers and hand shape, while also being capable of fitting the desired hand pose in the generated image. Given a generated failed image due to malformed hands, we utilize ControlNet modules to re-inject such correct hand information. Additionally, we uncover a phase transition phenomenon within ControlNet as we vary the control strength. It enables us to take advantage of more readily available synthetic data without suffering from the domain gap between realistic and synthetic hands. Experiments demonstrate that HandRefiner can significantly improve the generation quality quantitatively and qualitatively. The code is available at https://github.com/wenquanlu/HandRefiner .

Qi Zheng, Daqing Liu, Chaoyue Wang et al.

Vision-and-language navigation (VLN) simulates a visual agent that follows natural-language navigation instructions in real-world scenes. Existing approaches have made enormous progress in navigation in new environments, such as beam search, pre-exploration, and dynamic or hierarchical history encoding. To balance generalization and efficiency, we resort to memorizing visited scenarios apart from the ongoing route while navigating. In this work, we introduce a mechanism of Episodic Scene memory (ESceme) for VLN that wakes an agent's memories of past visits when it enters the current scene. The episodic scene memory allows the agent to envision a bigger picture of the next prediction. This way, the agent learns to utilize dynamically updated information instead of merely adapting to the current observations. We provide a simple yet effective implementation of ESceme by enhancing the accessible views at each location and progressively completing the memory while navigating. We verify the superiority of ESceme on short-horizon (R2R), long-horizon (R4R), and vision-and-dialog (CVDN) VLN tasks. Our ESceme also wins first place on the CVDN leaderboard. Code is available: \url{https://github.com/qizhust/esceme}.

Jinkai Tian, Xiaoyu Sun, Yuxuan Du et al.

Quantum computers are next-generation devices that hold promise to perform calculations beyond the reach of classical computers. A leading method towards achieving this goal is through quantum machine learning, especially quantum generative learning. Due to the intrinsic probabilistic nature of quantum mechanics, it is reasonable to postulate that quantum generative learning models (QGLMs) may surpass their classical counterparts. As such, QGLMs are receiving growing attention from the quantum physics and computer science communities, where various QGLMs that can be efficiently implemented on near-term quantum machines with potential computational advantages are proposed. In this paper, we review the current progress of QGLMs from the perspective of machine learning. Particularly, we interpret these QGLMs, covering quantum circuit born machines, quantum generative adversarial networks, quantum Boltzmann machines, and quantum autoencoders, as the quantum extension of classical generative learning models. In this context, we explore their intrinsic relation and their fundamental differences. We further summarize the potential applications of QGLMs in both conventional machine learning tasks and quantum physics. Last, we discuss the challenges and further research directions for QGLMs.

Minghui Hu, Chuanxia Zheng, Heliang Zheng et al.

The recently developed discrete diffusion models perform extraordinarily well in the text-to-image task, showing significant promise for handling the multi-modality signals. In this work, we harness these traits and present a unified multimodal generation model that can conduct both the "modality translation" and "multi-modality generation" tasks using a single model, performing text-based, image-based, and even vision-language simultaneous generation. Specifically, we unify the discrete diffusion process for multimodal signals by proposing a unified transition matrix. Moreover, we design a mutual attention module with fused embedding layer and a unified objective function to emphasise the inter-modal linkages, which are vital for multi-modality generation. Extensive experiments indicate that our proposed method can perform comparably to the state-of-the-art solutions in various generation tasks.

Ziqiang Li, Chaoyue Wang, Heliang Zheng et al.

Data-Efficient GANs (DE-GANs), which aim to learn generative models with a limited amount of training data, encounter several challenges for generating high-quality samples. Since data augmentation strategies have largely alleviated the training instability, how to further improve the generative performance of DE-GANs becomes a hotspot. Recently, contrastive learning has shown the great potential of increasing the synthesis quality of DE-GANs, yet related principles are not well explored. In this paper, we revisit and compare different contrastive learning strategies in DE-GANs, and identify (i) the current bottleneck of generative performance is the discontinuity of latent space; (ii) compared to other contrastive learning strategies, Instance-perturbation works towards latent space continuity, which brings the major improvement to DE-GANs. Based on these observations, we propose FakeCLR, which only applies contrastive learning on perturbed fake samples, and devises three related training techniques: Noise-related Latent Augmentation, Diversity-aware Queue, and Forgetting Factor of Queue. Our experimental results manifest the new state of the arts on both few-shot generation and limited-data generation. On multiple datasets, FakeCLR acquires more than 15% FID improvement compared to existing DE-GANs. Code is available at https://github.com/iceli1007/FakeCLR.

Yi Wu, Ziqiang Li, Chaoyue Wang et al.

In this study, we delve into the task of few-shot Generative Domain Adaptation (GDA), which involves transferring a pre-trained generator from one domain to a new domain using only a few reference images. Inspired by the way human brains acquire knowledge in new domains, we present an innovative generator structure called Domain Re-Modulation (DoRM). DoRM not only meets the criteria of high quality, large synthesis diversity, and cross-domain consistency, which were achieved by previous research in GDA, but also incorporates memory and domain association, akin to how human brains operate. Specifically, DoRM freezes the source generator and introduces new mapping and affine modules (M&A modules) to capture the attributes of the target domain during GDA. This process resembles the formation of new synapses in human brains. Consequently, a linearly combinable domain shift occurs in the style space. By incorporating multiple new M&A modules, the generator gains the capability to perform high-fidelity multi-domain and hybrid-domain generation. Moreover, to maintain cross-domain consistency more effectively, we introduce a similarity-based structure loss. This loss aligns the auto-correlation map of the target image with its corresponding auto-correlation map of the source image during training. Through extensive experiments, we demonstrate the superior performance of our DoRM and similarity-based structure loss in few-shot GDA, both quantitatively and qualitatively. The code will be available at https://github.com/wuyi2020/DoRM.

Haibin He, Xinyuan Chen, Chaoyue Wang et al.

Font generation is a difficult and time-consuming task, especially in those languages using ideograms that have complicated structures with a large number of characters, such as Chinese. To solve this problem, few-shot font generation and even one-shot font generation have attracted a lot of attention. However, most existing font generation methods may still suffer from (i) large cross-font gap challenge; (ii) subtle cross-font variation problem; and (iii) incorrect generation of complicated characters. In this paper, we propose a novel one-shot font generation method based on a diffusion model, named Diff-Font, which can be stably trained on large datasets. The proposed model aims to generate the entire font library by giving only one sample as the reference. Specifically, a large stroke-wise dataset is constructed, and a stroke-wise diffusion model is proposed to preserve the structure and the completion of each generated character. To our best knowledge, the proposed Diff-Font is the first work that developed diffusion models to handle the font generation task. The well-trained Diff-Font is not only robust to font gap and font variation, but also achieved promising performance on difficult character generation. Compared to previous font generation methods, our model reaches state-of-the-art performance both qualitatively and quantitatively.

Qi Zheng, Chaoyue Wang, Daqing Liu et al.

Multi-modal reasoning in visual question answering (VQA) has witnessed rapid progress recently. However, most reasoning models heavily rely on shortcuts learned from training data, which prevents their usage in challenging real-world scenarios. In this paper, we propose a simple but effective cross-modal contrastive learning strategy to get rid of the shortcut reasoning caused by imbalanced annotations and improve the overall performance. Different from existing contrastive learning with complex negative categories on coarse (Image, Question, Answer) triplet level, we leverage the correspondences between the language and image modalities to perform finer-grained cross-modal contrastive learning. We treat each Question-Answer (QA) pair as a whole, and differentiate between images that conform with it and those against it. To alleviate the issue of sampling bias, we further build connected graphs among images. For each positive pair, we regard the images from different graphs as negative samples and deduct the version of multi-positive contrastive learning. To our best knowledge, it is the first paper that reveals a general contrastive learning strategy without delicate hand-craft rules can contribute to robust VQA reasoning. Experiments on several mainstream VQA datasets demonstrate our superiority compared to the state of the arts. Code is available at \url{https://github.com/qizhust/cmcl_vqa_pl}.

Qi Zheng, Chaoyue Wang, Dadong Wang et al.

Concept learning constructs visual representations that are connected to linguistic semantics, which is fundamental to vision-language tasks. Although promising progress has been made, existing concept learners are still vulnerable to attribute perturbations and out-of-distribution compositions during inference. We ascribe the bottleneck to a failure of exploring the intrinsic semantic hierarchy of visual concepts, e.g. \{red, blue,...\} $\in$ `color' subspace yet cube $\in$ `shape'. In this paper, we propose a visual superordinate abstraction framework for explicitly modeling semantic-aware visual subspaces (i.e. visual superordinates). With only natural visual question answering data, our model first acquires the semantic hierarchy from a linguistic view, and then explores mutually exclusive visual superordinates under the guidance of linguistic hierarchy. In addition, a quasi-center visual concept clustering and a superordinate shortcut learning schemes are proposed to enhance the discrimination and independence of concepts within each visual superordinate. Experiments demonstrate the superiority of the proposed framework under diverse settings, which increases the overall answering accuracy relatively by 7.5\% on reasoning with perturbations and 15.6\% on compositional generalization tests.

Mengya Han, Heliang Zheng, Chaoyue Wang et al.

Few-shot part segmentation aims to separate different parts of an object given only a few annotated samples. Due to the challenge of limited data, existing works mainly focus on learning classifiers over pre-trained features, failing to learn task-specific features for part segmentation. In this paper, we propose to learn task-specific features in a "pre-training"-"fine-tuning" paradigm. We conduct prompt designing to reduce the gap between the pre-train task (i.e., image generation) and the downstream task (i.e., part segmentation), so that the GAN priors for generation can be leveraged for segmentation. This is achieved by projecting part segmentation maps into the RGB space and conducting interpolation between RGB segmentation maps and original images. Specifically, we design a fine-tuning strategy to progressively tune an image generator into a segmentation generator, where the supervision of the generator varying from images to segmentation maps by interpolation. Moreover, we propose a two-stream architecture, i.e., a segmentation stream to generate task-specific features, and an image stream to provide spatial constraints. The image stream can be regarded as a self-supervised auto-encoder, and this enables our model to benefit from large-scale support images. Overall, this work is an attempt to explore the internal relevance between generation tasks and perception tasks by prompt designing. Extensive experiments show that our model can achieve state-of-the-art performance on several part segmentation datasets.

Yiyang Chen, Shanshan Zhao, Changxing Ding et al.

In recent years, cross-modal domain adaptation has been studied on the paired 2D image and 3D LiDAR data to ease the labeling costs for 3D LiDAR semantic segmentation (3DLSS) in the target domain. However, in such a setting the paired 2D and 3D data in the source domain are still collected with additional effort. Since the 2D-3D projections can enable the 3D model to learn semantic information from the 2D counterpart, we ask whether we could further remove the need of source 3D data and only rely on the source 2D images. To answer it, this paper studies a new 3DLSS setting where a 2D dataset (source) with semantic annotations and a paired but unannotated 2D image and 3D LiDAR data (target) are available. To achieve 3DLSS in this scenario, we propose Cross-Modal and Cross-Domain Learning (CoMoDaL). Specifically, our CoMoDaL aims at modeling 1) inter-modal cross-domain distillation between the unpaired source 2D image and target 3D LiDAR data, and 2) the intra-domain cross-modal guidance between the target 2D image and 3D LiDAR data pair. In CoMoDaL, we propose to apply several constraints, such as point-to-pixel and prototype-to-pixel alignments, to associate the semantics in different modalities and domains by constructing mixed samples in two modalities. The experimental results on several datasets show that in the proposed setting, the developed CoMoDaL can achieve segmentation without the supervision of labeled LiDAR data. Ablations are also conducted to provide more analysis. Code will be available publicly.

Mengya Han, Heliang Zheng, Chaoyue Wang et al.

In this work, we address the task of few-shot part segmentation, which aims to segment the different parts of an unseen object using very few labeled examples. It is found that leveraging the textual space of a powerful pre-trained image-language model (such as CLIP) can be beneficial in learning visual features. Therefore, we develop a novel method termed PartSeg for few-shot part segmentation based on multimodal learning. Specifically, we design a part-aware prompt learning method to generate part-specific prompts that enable the CLIP model to better understand the concept of ``part'' and fully utilize its textual space. Furthermore, since the concept of the same part under different object categories is general, we establish relationships between these parts during the prompt learning process. We conduct extensive experiments on the PartImageNet and Pascal$\_$Part datasets, and the experimental results demonstrated that our proposed method achieves state-of-the-art performance.

Zhuqiang Lu, Kun Hu, Chaoyue Wang et al.

A 360-degree (omni-directional) image provides an all-encompassing spherical view of a scene. Recently, there has been an increasing interest in synthesising 360-degree images from conventional narrow field of view (NFoV) images captured by digital cameras and smartphones, for providing immersive experiences in various scenarios such as virtual reality. Yet, existing methods typically fall short in synthesizing intricate visual details or ensure the generated images align consistently with user-provided prompts. In this study, autoregressive omni-aware generative network (AOG-Net) is proposed for 360-degree image generation by out-painting an incomplete 360-degree image progressively with NFoV and text guidances joinly or individually. This autoregressive scheme not only allows for deriving finer-grained and text-consistent patterns by dynamically generating and adjusting the process but also offers users greater flexibility to edit their conditions throughout the generation process. A global-local conditioning mechanism is devised to comprehensively formulate the outpainting guidance in each autoregressive step. Text guidances, omni-visual cues, NFoV inputs and omni-geometry are encoded and further formulated with cross-attention based transformers into a global stream and a local stream into a conditioned generative backbone model. As AOG-Net is compatible to leverage large-scale models for the conditional encoder and the generative prior, it enables the generation to use extensive open-vocabulary text guidances. Comprehensive experiments on two commonly used 360-degree image datasets for both indoor and outdoor settings demonstrate the state-of-the-art performance of our proposed method. Our code will be made publicly available.

Ziqiang Li, Beihao Xia, Jing Zhang et al.

Generative Adversarial Networks (GANs) have achieved remarkable achievements in image synthesis. These successes of GANs rely on large scale datasets, requiring too much cost. With limited training data, how to stable the training process of GANs and generate realistic images have attracted more attention. The challenges of Data-Efficient GANs (DE-GANs) mainly arise from three aspects: (i) Mismatch Between Training and Target Distributions, (ii) Overfitting of the Discriminator, and (iii) Imbalance Between Latent and Data Spaces. Although many augmentation and pre-training strategies have been proposed to alleviate these issues, there lacks a systematic survey to summarize the properties, challenges, and solutions of DE-GANs. In this paper, we revisit and define DE-GANs from the perspective of distribution optimization. We conclude and analyze the challenges of DE-GANs. Meanwhile, we propose a taxonomy, which classifies the existing methods into three categories: Data Selection, GANs Optimization, and Knowledge Sharing. Last but not the least, we attempt to highlight the current problems and the future directions.

Ziqiang Li, Chaoyue Wang, Xue Rui et al.

Few-shot image generation aims to train generative models using a small number of training images. When there are few images available for training (e.g. 10 images), Learning From Scratch (LFS) methods often generate images that closely resemble the training data while Transfer Learning (TL) methods try to improve performance by leveraging prior knowledge from GANs pre-trained on large-scale datasets. However, current TL methods may not allow for sufficient control over the degree of knowledge preservation from the source model, making them unsuitable for setups where the source and target domains are not closely related. To address this, we propose a novel pipeline called Peer is your Pillar (PIP), which combines a target few-shot dataset with a peer dataset to create a data-unbalanced conditional generation. Our approach includes a class embedding method that separates the class space from the latent space, and we use a direction loss based on pre-trained CLIP to improve image diversity. Experiments on various few-shot datasets demonstrate the advancement of the proposed PIP, especially reduces the training requirements of few-shot image generation.

Liyao Tang, Zhe Chen, Shanshan Zhao et al.

Label-efficient segmentation aims to perform effective segmentation on input data using only sparse and limited ground-truth labels for training. This topic is widely studied in 3D point cloud segmentation due to the difficulty of annotating point clouds densely, while it is also essential for cost-effective segmentation on 2D images. Until recently, pseudo-labels have been widely employed to facilitate training with limited ground-truth labels, and promising progress has been witnessed in both the 2D and 3D segmentation. However, existing pseudo-labeling approaches could suffer heavily from the noises and variations in unlabelled data, which would result in significant discrepancies between generated pseudo-labels and current model predictions during training. We analyze that this can further confuse and affect the model learning process, which shows to be a shared problem in label-efficient learning across both 2D and 3D modalities. To address this issue, we propose a novel learning strategy to regularize the pseudo-labels generated for training, thus effectively narrowing the gaps between pseudo-labels and model predictions. More specifically, our method introduces an Entropy Regularization loss and a Distribution Alignment loss for label-efficient learning, resulting in an ERDA learning strategy. Interestingly, by using KL distance to formulate the distribution alignment loss, ERDA reduces to a deceptively simple cross-entropy-based loss which optimizes both the pseudo-label generation module and the segmentation model simultaneously. In addition, we innovate in the pseudo-label generation to make our ERDA consistently effective across both 2D and 3D data modalities for segmentation. Enjoying simplicity and more modality-agnostic pseudo-label generation, our method has shown outstanding performance in fully utilizing all unlabeled data points for training across ...

Xingyu Yang, Daqing Liu, Heng Zhang et al.

Composed image retrieval is a type of image retrieval task where the user provides a reference image as a starting point and specifies a text on how to shift from the starting point to the desired target image. However, most existing methods focus on the composition learning of text and reference images and oversimplify the text as a description, neglecting the inherent structure and the user's shifting intention of the texts. As a result, these methods typically take shortcuts that disregard the visual cue of the reference images. To address this issue, we reconsider the text as instructions and propose a Semantic Shift network (SSN) that explicitly decomposes the semantic shifts into two steps: from the reference image to the visual prototype and from the visual prototype to the target image. Specifically, SSN explicitly decomposes the instructions into two components: degradation and upgradation, where the degradation is used to picture the visual prototype from the reference image, while the upgradation is used to enrich the visual prototype into the final representations to retrieve the desired target image. The experimental results show that the proposed SSN demonstrates a significant improvement of 5.42% and 1.37% on the CIRR and FashionIQ datasets, respectively, and establishes a new state-of-the-art performance. Codes will be publicly available.

Qi Zheng, Chaoyue Wang, Dadong Wang

Image paragraph captioning aims to describe a given image with a sequence of coherent sentences. Most existing methods model the coherence through the topic transition that dynamically infers a topic vector from preceding sentences. However, these methods still suffer from immediate or delayed repetitions in generated paragraphs because (i) the entanglement of syntax and semantics distracts the topic vector from attending pertinent visual regions; (ii) there are few constraints or rewards for learning long-range transitions. In this paper, we propose a bypass network that separately models semantics and linguistic syntax of preceding sentences. Specifically, the proposed model consists of two main modules, i.e. a topic transition module and a sentence generation module. The former takes previous semantic vectors as queries and applies attention mechanism on regional features to acquire the next topic vector, which reduces immediate repetition by eliminating linguistics. The latter decodes the topic vector and the preceding syntax state to produce the following sentence. To further reduce delayed repetition in generated paragraphs, we devise a replacement-based reward for the REINFORCE training. Comprehensive experiments on the widely used benchmark demonstrate the superiority of the proposed model over the state of the art for coherence while maintaining high accuracy.

Ziqiang Li, Yi Wu, Chaoyue Wang et al.

3D-aware image generation necessitates extensive training data to ensure stable training and mitigate the risk of overfitting. This paper first considers a novel task known as One-shot 3D Generative Domain Adaptation (GDA), aimed at transferring a pre-trained 3D generator from one domain to a new one, relying solely on a single reference image. One-shot 3D GDA is characterized by the pursuit of specific attributes, namely, high fidelity, large diversity, cross-domain consistency, and multi-view consistency. Within this paper, we introduce 3D-Adapter, the first one-shot 3D GDA method, for diverse and faithful generation. Our approach begins by judiciously selecting a restricted weight set for fine-tuning, and subsequently leverages four advanced loss functions to facilitate adaptation. An efficient progressive fine-tuning strategy is also implemented to enhance the adaptation process. The synergy of these three technological components empowers 3D-Adapter to achieve remarkable performance, substantiated both quantitatively and qualitatively, across all desired properties of 3D GDA. Furthermore, 3D-Adapter seamlessly extends its capabilities to zero-shot scenarios, and preserves the potential for crucial tasks such as interpolation, reconstruction, and editing within the latent space of the pre-trained generator. Code will be available at https://github.com/iceli1007/3D-Adapter.

Yongshun Zhang, Zhongyi Fan, Yonghang Zhang et al.

In recent years, large-scale generative models for visual content (\textit{e.g.,} images, videos, and 3D objects/scenes) have made remarkable progress. However, training large-scale video generation models remains particularly challenging and resource-intensive due to cross-modal text-video alignment, the long sequences involved, and the complex spatiotemporal dependencies. To address these challenges, we present a training framework that optimizes four pillars: (i) data processing, (ii) model architecture, (iii) training strategy, and (iv) infrastructure for large-scale video generation models. These optimizations delivered significant efficiency gains and performance improvements across all stages of data preprocessing, video compression, parameter scaling, curriculum-based pretraining, and alignment-focused post-training. Our resulting model, MUG-V 10B, matches recent state-of-the-art video generators overall and, on e-commerce-oriented video generation tasks, surpasses leading open-source baselines in human evaluations. More importantly, we open-source the complete stack, including model weights, Megatron-Core-based large-scale training code, and inference pipelines for video generation and enhancement. To our knowledge, this is the first public release of large-scale video generation training code that exploits Megatron-Core to achieve high training efficiency and near-linear multi-node scaling, details are available in https://github.com/Shopee-MUG/MUG-V.

Yue He, Chen Chen, Jing Zhang et al.

Existing Scene Text Recognition (STR) methods typically use a language model to optimize the joint probability of the 1D character sequence predicted by a visual recognition (VR) model, which ignore the 2D spatial context of visual semantics within and between character instances, making them not generalize well to arbitrary shape scene text. To address this issue, we make the first attempt to perform textual reasoning based on visual semantics in this paper. Technically, given the character segmentation maps predicted by a VR model, we construct a subgraph for each instance, where nodes represent the pixels in it and edges are added between nodes based on their spatial similarity. Then, these subgraphs are sequentially connected by their root nodes and merged into a complete graph. Based on this graph, we devise a graph convolutional network for textual reasoning (GTR) by supervising it with a cross-entropy loss. GTR can be easily plugged in representative STR models to improve their performance owing to better textual reasoning. Specifically, we construct our model, namely S-GTR, by paralleling GTR to the language model in a segmentation-based STR baseline, which can effectively exploit the visual-linguistic complementarity via mutual learning. S-GTR sets new state-of-the-art on six challenging STR benchmarks and generalizes well to multi-linguistic datasets. Code is available at https://github.com/adeline-cs/GTR.

Jiuding Yang, Weidong Guo, Bang Liu et al.

Although conceptualization has been widely studied in semantics and knowledge representation, it is still challenging to find the most accurate concept phrases to characterize the main idea of a text snippet on the fast-growing social media. This is partly attributed to the fact that most knowledge bases contain general terms of the world, such as trees and cars, which do not have the defining power or are not interesting enough to social media app users. Another reason is that the intricacy of natural language allows the use of tense, negation and grammar to change the logic or emphasis of language, thus conveying completely different meanings. In this paper, we present TAG, a high-quality concept matching dataset consisting of 10,000 labeled pairs of fine-grained concepts and web-styled natural language sentences, mined from the open-domain social media. The concepts we consider represent the trending interests of online users. Associated with TAG is a concept graph of these fine-grained concepts and entities to provide the structural context information. We evaluate a wide range of popular neural text matching models as well as pre-trained language models on TAG, and point out their insufficiency to tag social media content with the most appropriate concept. We further propose a novel graph-graph matching method that demonstrates superior abstraction and generalization performance by better utilizing both the structural context in the concept graph and logic interactions between semantic units in the sentence via syntactic dependency parsing. We open-source both the TAG dataset and the proposed methods to facilitate further research.

Youjian Zhang, Chaoyue Wang, Stephen J. Maybank et al.

Motion blur in dynamic scenes is an important yet challenging research topic. Recently, deep learning methods have achieved impressive performance for dynamic scene deblurring. However, the motion information contained in a blurry image has yet to be fully explored and accurately formulated because: (i) the ground truth of dynamic motion is difficult to obtain; (ii) the temporal ordering is destroyed during the exposure; and (iii) the motion estimation from a blurry image is highly ill-posed. By revisiting the principle of camera exposure, motion blur can be described by the relative motions of sharp content with respect to each exposed position. In this paper, we define exposure trajectories, which represent the motion information contained in a blurry image and explain the causes of motion blur. A novel motion offset estimation framework is proposed to model pixel-wise displacements of the latent sharp image at multiple timepoints. Under mild constraints, our method can recover dense, (non-)linear exposure trajectories, which significantly reduce temporal disorder and ill-posed problems. Finally, experiments demonstrate that the recovered exposure trajectories not only capture accurate and interpretable motion information from a blurry image, but also benefit motion-aware image deblurring and warping-based video extraction tasks. Codes are available on https://github.com/yjzhang96/Motion-ETR.

Yi Wu, Ziqiang Li, Heliang Zheng et al.

Drawing on recent advancements in diffusion models for text-to-image generation, identity-preserved personalization has made significant progress in accurately capturing specific identities with just a single reference image. However, existing methods primarily integrate reference images within the text embedding space, leading to a complex entanglement of image and text information, which poses challenges for preserving both identity fidelity and semantic consistency. To tackle this challenge, we propose Infinite-ID, an ID-semantics decoupling paradigm for identity-preserved personalization. Specifically, we introduce identity-enhanced training, incorporating an additional image cross-attention module to capture sufficient ID information while deactivating the original text cross-attention module of the diffusion model. This ensures that the image stream faithfully represents the identity provided by the reference image while mitigating interference from textual input. Additionally, we introduce a feature interaction mechanism that combines a mixed attention module with an AdaIN-mean operation to seamlessly merge the two streams. This mechanism not only enhances the fidelity of identity and semantic consistency but also enables convenient control over the styles of the generated images. Extensive experimental results on both raw photo generation and style image generation demonstrate the superior performance of our proposed method.

Wenjie Xuan, Yufei Xu, Shanshan Zhao et al.

ControlNet excels at creating content that closely matches precise contours in user-provided masks. However, when these masks contain noise, as a frequent occurrence with non-expert users, the output would include unwanted artifacts. This paper first highlights the crucial role of controlling the impact of these inexplicit masks with diverse deterioration levels through in-depth analysis. Subsequently, to enhance controllability with inexplicit masks, an advanced Shape-aware ControlNet consisting of a deterioration estimator and a shape-prior modulation block is devised. The deterioration estimator assesses the deterioration factor of the provided masks. Then this factor is utilized in the modulation block to adaptively modulate the model's contour-following ability, which helps it dismiss the noise part in the inexplicit masks. Extensive experiments prove its effectiveness in encouraging ControlNet to interpret inaccurate spatial conditions robustly rather than blindly following the given contours, suitable for diverse kinds of conditions. We showcase application scenarios like modifying shape priors and composable shape-controllable generation. Codes are available at github.

Jing Zhao, Heliang Zheng, Chaoyue Wang et al.

Large-scale text-to-image diffusion models, (e.g., DALL-E, SDXL) are capable of generating famous persons by simply referring to their names. Is it possible to make such models generate generic identities as simple as the famous ones, e.g., just use a name? In this paper, we explore the existence of a "Name Space", where any point in the space corresponds to a specific identity. Fortunately, we find some clues in the feature space spanned by text embedding of celebrities' names. Specifically, we first extract the embeddings of celebrities' names in the Laion5B dataset with the text encoder of diffusion models. Such embeddings are used as supervision to learn an encoder that can predict the name (actually an embedding) of a given face image. We experimentally find that such name embeddings work well in promising the generated image with good identity consistency. Note that like the names of celebrities, our predicted name embeddings are disentangled from the semantics of text inputs, making the original generation capability of text-to-image models well-preserved. Moreover, by simply plugging such name embeddings, all variants (e.g., from Civitai) derived from the same base model (i.e., SDXL) readily become identity-aware text-to-image models. Project homepage: \url{https://magicfusion.github.io/MagicNaming/}.

Jing Zhao, Heliang Zheng, Chaoyue Wang et al.

Classifier-free guidance is an effective sampling technique in diffusion models that has been widely adopted. The main idea is to extrapolate the model in the direction of text guidance and away from null-text guidance. In this paper, we demonstrate that null-text guidance in diffusion models is secretly a cartoon-style creator, i.e., the generated images can be efficiently transformed into cartoons by simply perturbing the null-text guidance. Specifically, we proposed two disturbance methods, i.e., Rollback disturbance (Back-D) and Image disturbance (Image-D), to construct misalignment between the noisy images used for predicting null-text guidance and text guidance (subsequently referred to as \textbf{null-text noisy image} and \textbf{text noisy image} respectively) in the sampling process. Back-D achieves cartoonization by altering the noise level of null-text noisy image via replacing $x_t$ with $x_{t+Δt}$. Image-D, alternatively, produces high-fidelity, diverse cartoons by defining $x_t$ as a clean input image, which further improves the incorporation of finer image details. Through comprehensive experiments, we delved into the principle of noise disturbing for null-text and uncovered that the efficacy of disturbance depends on the correlation between the null-text noisy image and the source image. Moreover, our proposed techniques, which can generate cartoon images and cartoonize specific ones, are training-free and easily integrated as a plug-and-play component in any classifier-free guided diffusion model. Project page is available at \url{https://nulltextforcartoon.github.io/}.

Jianbin Zheng, Daqing Liu, Chaoyue Wang et al.

Existing multimodal conditional image synthesis (MCIS) methods generate images conditioned on any combinations of various modalities that require all of them must be exactly conformed, hindering the synthesis controllability and leaving the potential of cross-modality under-exploited. To this end, we propose to generate images conditioned on the compositions of multimodal control signals, where modalities are imperfectly complementary, i.e., composed multimodal conditional image synthesis (CMCIS). Specifically, we observe two challenging issues of the proposed CMCIS task, i.e., the modality coordination problem and the modality imbalance problem. To tackle these issues, we introduce a Mixture-of-Modality-Tokens Transformer (MMoT) that adaptively fuses fine-grained multimodal control signals, a multimodal balanced training loss to stabilize the optimization of each modality, and a multimodal sampling guidance to balance the strength of each modality control signal. Comprehensive experimental results demonstrate that MMoT achieves superior performance on both unimodal conditional image synthesis (UCIS) and MCIS tasks with high-quality and faithful image synthesis on complex multimodal conditions. The project website is available at https://jabir-zheng.github.io/MMoT.

Hao Guan, Chaoyue Wang, Dacheng Tao

Mild cognitive impairment (MCI) conversion prediction, i.e., identifying MCI patients of high risks converting to Alzheimer's disease (AD), is essential for preventing or slowing the progression of AD. Although previous studies have shown that the fusion of multi-modal data can effectively improve the prediction accuracy, their applications are largely restricted by the limited availability or high cost of multi-modal data. Building an effective prediction model using only magnetic resonance imaging (MRI) remains a challenging research topic. In this work, we propose a multi-modal multi-instance distillation scheme, which aims to distill the knowledge learned from multi-modal data to an MRI-based network for MCI conversion prediction. In contrast to existing distillation algorithms, the proposed multi-instance probabilities demonstrate a superior capability of representing the complicated atrophy distributions, and can guide the MRI-based network to better explore the input MRI. To our best knowledge, this is the first study that attempts to improve an MRI-based prediction model by leveraging extra supervision distilled from multi-modal information. Experiments demonstrate the advantage of our framework, suggesting its potentials in the data-limited clinical settings.

Hao Xiong, Chaoyue Wang, Dacheng Tao et al.

Multiple sclerosis (MS) is an inflammatory demyelinating disease of the central nervous system (CNS) that results in focal injury to the grey and white matter. The presence of white matter lesions biases morphometric analyses such as registration, individual longitudinal measurements and tissue segmentation for brain volume measurements. Lesion-inpainting with intensities derived from surrounding healthy tissue represents one approach to alleviate such problems. However, existing methods inpaint lesions based on texture information derived from local surrounding tissue, often leading to inconsistent inpainting and the generation of artifacts such as intensity discrepancy and blurriness. Based on these observations, we propose non-local partial convolutions (NLPC) that integrates a Unet-like network with the non-local module. The non-local module is exploited to capture long range dependencies between the lesion area and remaining normal-appearing brain regions. Then, the lesion area is filled by referring to normal-appearing regions with more similar features. This method generates inpainted regions that appear more realistic and natural. Our quantitative experimental results also demonstrate superiority of this technique of existing state-of-the-art inpainting methods.

Chaoyue Wang, Chang Xu, Xin Yao et al.

Generative adversarial networks (GAN) have been effective for learning generative models for real-world data. However, existing GANs (GAN and its variants) tend to suffer from training problems such as instability and mode collapse. In this paper, we propose a novel GAN framework called evolutionary generative adversarial networks (E-GAN) for stable GAN training and improved generative performance. Unlike existing GANs, which employ a pre-defined adversarial objective function alternately training a generator and a discriminator, we utilize different adversarial training objectives as mutation operations and evolve a population of generators to adapt to the environment (i.e., the discriminator). We also utilize an evaluation mechanism to measure the quality and diversity of generated samples, such that only well-performing generator(s) are preserved and used for further training. In this way, E-GAN overcomes the limitations of an individual adversarial training objective and always preserves the best offspring, contributing to progress in and the success of GANs. Experiments on several datasets demonstrate that E-GAN achieves convincing generative performance and reduces the training problems inherent in existing GANs.

Chaoyue Wang, Chang Xu, Chaohui Wang et al.

In this paper, we propose a principled Perceptual Adversarial Networks (PAN) for image-to-image transformation tasks. Unlike existing application-specific algorithms, PAN provides a generic framework of learning mapping relationship between paired images (Fig. 1), such as mapping a rainy image to its de-rained counterpart, object edges to its photo, semantic labels to a scenes image, etc. The proposed PAN consists of two feed-forward convolutional neural networks (CNNs), the image transformation network T and the discriminative network D. Through combining the generative adversarial loss and the proposed perceptual adversarial loss, these two networks can be trained alternately to solve image-to-image transformation tasks. Among them, the hidden layers and output of the discriminative network D are upgraded to continually and automatically discover the discrepancy between the transformed image and the corresponding ground-truth. Simultaneously, the image transformation network T is trained to minimize the discrepancy explored by the discriminative network D. Through the adversarial training process, the image transformation network T will continually narrow the gap between transformed images and ground-truth images. Experiments evaluated on several image-to-image transformation tasks (e.g., image de-raining, image inpainting, etc.) show that the proposed PAN outperforms many related state-of-the-art methods.