Talking-head video editing aims to efficiently insert, delete, and substitute the word of a pre-recorded video through a text transcript editor. The key challenge for this task is obtaining an editing model that generates new talking-head video clips which simultaneously have accurate lip synchronization and motion smoothness. Previous approaches, including 3DMM-based (3D Morphable Model) methods and NeRF-based (Neural Radiance Field) methods, are sub-optimal in that they either require minutes of source videos and days of training time or lack the disentangled control of verbal (e.g., lip motion) and non-verbal (e.g., head pose and expression) representations for video clip insertion. In this work, we fully utilize the video context to design a novel framework for talking-head video editing, which achieves efficiency, disentangled motion control, and sequential smoothness. Specifically, we decompose this framework to motion prediction and motion-conditioned rendering: (1) We first design an animation prediction module that efficiently obtains smooth and lip-sync motion sequences conditioned on the driven speech. This module adopts a non-autoregressive network to obtain context prior and improve the prediction efficiency, and it learns a speech-animation mapping prior with better generalization to novel speech from a multi-identity video dataset. (2) We then introduce a neural rendering module to synthesize the photo-realistic and full-head video frames given the predicted motion sequence. This module adopts a pre-trained head topology and uses only few frames for efficient fine-tuning to obtain a person-specific rendering model. Extensive experiments demonstrate that our method efficiently achieves smoother editing results with higher image quality and lip accuracy using less data than previous methods.
论文检索
输入标题、作者或关键词，从 8,216 篇学术成果中精准定位
Single image dehazing is a challenging task that requires both local detail and global distribution, and can be applied to various scenarios. However, physics-based dehazing algorithms perform well only in specific settings, while CNN-based algorithms struggle with capturing global information, and ViT-based approaches suffer from inadequate representation of local details. The shortcomings of the above three types of methods lead to issues such as imbalanced colors and incoherent details in the predicted haze-free image. To address these challenges, we propose a new Low-cost Hybrid Network called LHNet. The key insight of LHNet is the effective hybrid of different features, which can achieve better information fusion in the form of feature awareness at the cost of few parameters. This fusion approach narrows the gap between different features and enables LHNet to autonomously choose the fusion granularity to maximize the utilization of prior, local and global information. Extensive experiments are performed on the mainstream dehazing datasets, and the results show that LHNet achieves state-of-the-art performance in single image dehazing. By adopting our fusion approach, a better dehazing effect can be achieved than with other dehazing algorithms with more parameters, even when only CNN and ViT are used. The code is available at https://github.com/SHYuanBest/LHNet.
Egocentric Action Recognition (EAR) is required to recognize both the interacting objects (noun) and the motion (verb) against cluttered backgrounds with distracting objects. For capturing interacting objects, traditional approaches heavily rely on luxury object annotations or detectors, though a few works heuristically enumerate the fixed sets of verb-constrained prototypes to roughly exclude the background. For capturing motion, the inherent variations of motion duration among egocentric videos with different lengths are almost ignored. To this end, we propose a novel Slowfast Diversity-aware Prototype learning (SDP) to effectively capture interacting objects by learning compact yet diverse prototypes, and adaptively capture motion in either long-time video or short-time video. Specifically, we present a new Part-to-Prototype (P2P) scheme to learn prototypes from raw videos covering the interacting objects by refining the semantic information from part level to prototype level. Moreover, for adaptively capturing motion, we design a new Slow-Fast Context (SFC) mechanism that explores the Up/Down augmentations for the prototype representation at the semantic level to strengthen the transient dynamic information in short-time videos and eliminate the redundant dynamic information in long-time videos, which are further fine-complemented via the slow- and fast-aware attentions. Extensive experiments demonstrate SDP outperforms state-of-the-art methods on two large-scale egocentric video benchmarks, i.e., EPIC-KITCHENS-100 and EGTEA.
The task of spatio-temporal video grounding involves identifying the spatial and temporal regions in a video that correspond to the objects or actions described in a given textual description. However, current models used for spatio-temporal video grounding often rely heavily on spatio-temporal priors to make the predictions. As a result, they may suffer from spurious correlations and lack the ability to generalize well to new or diverse scenarios. To overcome this limitation, we introduce a deconfounded multimodal learning framework, which utilizes a structural causal model to treat dataset biases as a confounder and subsequently remove their confounding effect. Through this framework, we can perform causal intervention on the multimodal input and derive an unbiased estimation formula through the do-calculus technique. In order to tackle the challenge of diverse and often unobservable confounders, we further propose a novel retrieval-based approach with a causal mask mechanism. The proposed method leverages analogical reasoning to facilitate deconfounded learning and mitigate dataset biases, enabling unbiased spatio-temporal prediction without explicitly modeling the confounding factors. Extensive experiments on two challenging benchmarks have well verified the effectiveness and rationality of our proposed solution.
Unsupervised domain adaptation (UDA) involves predicting unlabeled data in a target domain by using labeled data from the source domain. However, recent advances in pseudo-labeling (PL) methods have been hampered by noisy pseudo-labels that diminish the local discriminativeness of the target structure. Although neighborhood-based PL can help preserve the local structure, it also risks assigning the whole local neighborhood to the wrong semantic category. To address this issue, we propose a novel framework called neighborhood consistency learning (NCL) that operates at both the semantic and instance levels and features a new consistency objective function. Specifically, our objective function aims to promote semantic consistency in the target neighborhood by computing the correlation matrix between the target samples and their neighborhood aggregation over a batch and matching the correlation matrix to an identity matrix. Importantly, our approach allows the target neighborhood to receive gradients from several potential positive categories instead of just one certain category. Our extensive experiments on UDA benchmarks demonstrate the effectiveness of NCL over other state-of-the-art PL-based methods.
Personalized fashion matching task aims to predict the compatible fashion items given available ones for specific users through the effective modeling of the third-order interaction patterns among the user and item pairs. To achieve this, previous methods separately model two key components, user-item and item-item relationships, which ignore the inherent correlations between them and lead to undesirable performance. With a new perspective, this paper proposes to formulate the personalized item matching as the multi-relational connectivity and apply a single-component translation operation to model the targeted third-order interactions. With user-item-item interactions naturally constructing a multi-relational graph, we further device two graph learning modules to enhance the translation-based matching approach from two perspectives,C ontext and Path. The proposed method, named CP-TransMatch, has been tested with extensive experiments on three benchmark fashion datasets and proven effective. It sets the new SOTA for the personalized fashion matching task.
3D content creation via text-driven stylization has played a fundamental challenge to multimedia and graphics community. Recent advances of cross-modal foundation models (e.g., CLIP) have made this problem feasible. Those approaches commonly leverage CLIP to align the holistic semantics of stylized mesh with the given text prompt. Nevertheless, it is not trivial to enable more controllable stylization of fine-grained details in 3D meshes solely based on such semantic-level cross-modal supervision. In this work, we propose a new 3DStyle-Diffusion model that triggers fine-grained stylization of 3D meshes with additional controllable appearance and geometric guidance from 2D Diffusion models. Technically, 3DStyle-Diffusion first parameterizes the texture of 3D mesh into reflectance properties and scene lighting using implicit MLP networks. Meanwhile, an accurate depth map of each sampled view is achieved conditioned on 3D mesh. Then, 3DStyle-Diffusion leverages a pre-trained controllable 2D Diffusion model to guide the learning of rendered images, encouraging the synthesized image of each view semantically aligned with text prompt and geometrically consistent with depth map. This way elegantly integrates both image rendering via implicit MLP networks and diffusion process of image synthesis in an end-to-end fashion, enabling a high-quality fine-grained stylization of 3D meshes. We also build a new dataset derived from Objaverse and the evaluation protocol for this task. Through both qualitative and quantitative experiments, we validate the capability of our 3DStyle-Diffusion. Source code and data are available at https://github.com/yanghb22-fdu/3DStyle-Diffusion-Official.
An ideal multi-modal recommendation system is supposed to be timely updated with the latest modality information and interaction data because the distribution discrepancy between new data and historical data will lead to severe recommendation performance deterioration. However, upgrading a recommendation system with numerous new data consumes much time and computing resources. To mitigate this problem, we propose a Task-Adversarial Adaptation (TAA) framework, which is able to align data distributions and reduce resource consumption at the same time. This framework is specifically designed to align distributions of embedded features for different recommendation tasks between the source domain (i.e., historical data) and the target domain (i.e., new data). Technically, we design a domain feature discriminator for each task to distinguish which domain a feature comes from. By the two-player min-max game between the feature discriminator and the feature embedding network, the feature embedding network is able to align the source and target data distributions. With the ability to align source and target distributions, we are able to reduce the number of training samples by random sampling. In addition, we formulate the proposed approach as a plug-and-play module to accelerate the model training and improve the performance of mainstream multi-modal multi-task recommendation systems. We evaluate our method by predicting the Click-Through Rate (CTR) in e-commerce scenarios. Extensive experiments verify that our method is able to significantly improve prediction performance and accelerate model training on the target domain. For instance, our method is able to surpass the previous state-of-the-art method by 2.45% in terms of Area Under Curve (AUC) on AliExpress_US dataset while only utilizing one percent of the target data in training. Code: https://github.com/TL-UESTC/TAA.
Taxonomy expansion task is essential in organizing the ever-increasing volume of new concepts into existing taxonomies. Most existing methods focus exclusively on using textual semantics, leading to an inability to generalize to unseen terms and the "Prototypical Hypernym Problem." In this paper, we propose Visual Taxonomy Expansion (VTE), introducing visual features into the taxonomy expansion task. We propose a textual hypernymy learning task and a visual prototype learning task to cluster textual and visual semantics. In addition to the tasks on respective modalities, we introduce a hyper-proto constraint that integrates textual and visual semantics to produce fine-grained visual semantics. Our method is evaluated on two datasets, where we obtain compelling results. Specifically, on the Chinese taxonomy dataset, our method significantly improves accuracy by 8.75%. Additionally, our approach performs better than ChatGPT on the Chinese taxonomy dataset.
The cold-start recommendation has been one of the most central problems in online platforms where new users or items arrive continuously. Although existing meta-learning based models with globally sharing knowledge show good performance in most cold-start scenarios, the ability to handle challenges on intention heterogeneity and prediction uncertainty is missing, and these two challenges are particularly evident in cold-start scenarios with fewer interaction data. To tackle the above challenges, in this paper, we present an uncertainty-aware Stochastic Meta Process with Doubly Intention learning (DISMP) for the cold-start recommendation, which has promising properties in uncertainty quantification. With the aid of the meta-learning stochastic process, DISMP can store general knowledge by capturing the relevance of different user-item pairs in terms of intentions and concepts, which is capable of rapid adaptation to new users and items. Furthermore, intentions with general and specific levels are extracted by doubly distinguishing the role of latent variables, which is able to capture the dependencies across different types of intentions and concepts. Empirical results show that our approach can achieve substantial improvement over the state-of-the-art baselines on cold-start recommendations with different perspectives.
Temporal recommendation systems aim to suggest items to users at the optimal time. However, the significant imbalance of items in the training data poses a major challenge to predictive accuracy. Existing approaches attempt to alleviate this issue by modifying the loss function or utilizing resampling techniques, but such approaches may inadvertently amplify the specificity of certain behaviors. To address this problem, we propose a novel temporal recommendation algorithm called HCRec. The algorithm adopts a transformer-like architecture, focuses on decoupling variant and invariant information according to hierarchical categories, and then uses invariant prototypes to mitigate the impact of special behaviors. Although category-enhanced recommendation is not a new concept, current approaches have considerable limitations. They usually incorporate the category information into a black-box model as another input sequence, which can easily lead to overfitting. In contrast, HCRec creates a hierarchical relationship tree of items and categories, and establishes mapping rules to classify nodes as variant or invariant. Then, an attention model is used to extract potential variant node features from historical behaviors and enhance them with external knowledge. The invariant node features are fused based on the relationship tree, and finally used for recommendation. On top of this, two time kernel functions are designed to simulate short-term and long-period time effects. Experiments on real-world datasets show that hierarchical category-enhanced prototype learning helps alleviate the imbalance problem, enabling HCRec to outperform all baseline methods.
Automatic detection of multimodal fake news has gained a widespread attention recently. Many existing approaches seek to fuse unimodal features to produce multimodal news representations. However, the potential of powerful cross-modal contrastive learning methods for fake news detection has not been well exploited. Besides, how to aggregate features from different modalities to boost the performance of the decision-making process is still an open question. To address that, we propose COOLANT, a cross-modal contrastive learning framework for multimodal fake news detection, aiming to achieve more accurate image-text alignment. To further capture the fine-grained alignment between vision and language, we leverage an auxiliary task to soften the loss term of negative samples during the contrast process. A cross-modal fusion module is developed to learn the cross-modality correlations. An attention mechanism with an attention guidance module is implemented to help effectively and interpretably aggregate the aligned unimodal representations and the cross-modality correlations. Finally, we evaluate the COOLANT and conduct a comparative study on two widely used datasets, Twitter and Weibo. The experimental results demonstrate that our COOLANT outperforms previous approaches by a large margin and achieves new state-of-the-art results on the two datasets.
Human-Object Interaction (HOI) detection aims to understand the interactions between humans and objects, which plays a curtail role in high-level semantic understanding tasks. However, most works pursue designing better architectures to learn overall features more efficiently, while ignoring the long-tail nature of interaction-object pair categories. In this paper, we propose to alleviate the impact of such an unbalanced distribution via Virtual Image Leaning (VIL). Firstly, a novel label-to-image approach, Multiple Steps Image Creation (MUSIC), is proposed to create a high-quality dataset that has a consistent distribution with real images. In this stage, virtual images are generated based on prompts with specific characterizations and selected by multi-filtering processes. Secondly, we use both virtual and real images to train the model with the teacher-student framework. Considering the initial labels of some virtual images are inaccurate and inadequate, we devise an Adaptive Matching-and-Filtering (AMF) module to construct pseudo-labels. Our method is independent of the internal structure of HOI detectors, so it can be combined with off-the-shelf methods by training merely 10 additional epochs. With the assistance of our method, multiple methods obtain significant improvements, and new state-of-the-art results are achieved on two benchmarks.
With the recent progress in large-scale vision and language representation learning, Vision Language Pre-training (VLP) models have achieved promising improvements on various multi-modal downstream tasks. Albeit powerful, these models have not fully leveraged world knowledge to their advantage. A key challenge of knowledge-augmented VLP is the lack of clear connections between knowledge and multi-modal data. Moreover, not all knowledge present in images/texts is useful, therefore prior approaches often struggle to effectively integrate knowledge, visual, and textual information. In this study, we propose REtrieval-based knowledge Augmented Vision Language (REAVL), a novel knowledge-augmented pre-training framework to address the above issues. For the first time, we introduce a knowledge-aware self-supervised learning scheme that efficiently establishes the correspondence between knowledge and multi-modal data and identifies informative knowledge to improve the modeling of alignment and interactions between visual and textual modalities. By adaptively integrating informative knowledge with visual and textual information, REAVL achieves new state-of-the-art performance uniformly on knowledge-based vision-language understanding and multi-modal entity linking tasks, as well as competitive results on general vision-language tasks while only using 0.2% pre-training data of the best models. Our model shows strong sample efficiency and effective knowledge utilization.
Source-free object detection (SFOD) aims to adapt a source-trained detector to an unlabeled target domain without access to the labeled source data. Current SFOD methods utilize a threshold-based pseudo-label approach in the adaptation phase, which is typically limited to high-confidence pseudo-labels and results in a loss of information. To address this issue, we propose a new approach to take full advantage of pseudo-labels by introducing high and low confidence thresholds. Specifically, the pseudo-labels with confidence scores above the high threshold are used conventionally, while those between the low and high thresholds are exploited using the Low-confidence Pseudo-labels Utilization (LPU) module. The LPU module consists of Proposal Soft Training (PST) and Local Spatial Contrastive Learning (LSCL). PST generates soft labels of proposals for soft training, which can mitigate the label mismatch problem. LSCL exploits the local spatial relationship of proposals to improve the model's ability to differentiate between spatially adjacent proposals, thereby optimizing representational features further. Combining the two components overcomes the challenges faced by traditional methods in utilizing low-confidence pseudo-labels. Extensive experiments on five cross-domain object detection benchmarks demonstrate that our proposed method outperforms the previous SFOD methods, achieving state-of-the-art performance.
Language-guided human motion synthesis has been a challenging task due to the inherent complexity and diversity of human behaviors. Previous methods face limitations in generalization to novel actions, often resulting in unrealistic or incoherent motion sequences. In this paper, we propose ATOM (ATomic mOtion Modeling) to mitigate this problem, by decomposing actions into atomic actions, and employing a curriculum learning strategy to learn atomic action composition. First, we disentangle complex human motions into a set of atomic actions during learning, and then assemble novel actions using the learned atomic actions, which offers better adaptability to new actions. Moreover, we introduce a curriculum learning training strategy that leverages masked motion modeling with a gradual increase in the mask ratio, and thus facilitates atomic action assembly. This approach mitigates the overfitting problem commonly encountered in previous methods while enforcing the model to learn better motion representations. We demonstrate the effectiveness of ATOM through extensive experiments, including text-to-motion and action-to-motion synthesis tasks. We further illustrate its superiority in synthesizing plausible and coherent text-guided human motion sequences.
Dual-Modal Attention-Enhanced Text-Video Retrieval with Triplet Partial Margin Contrastive Learning
PDF ↗In recent years, the explosion of web videos makes text-video retrieval increasingly essential and popular for video filtering, recommendation, and search. Text-video retrieval aims to rank relevant text/video higher than irrelevant ones. The core of this task is to precisely measure the cross-modal similarity between texts and videos. Recently, contrastive learning methods have shown promising results for text-video retrieval, most of which focus on the construction of positive and negative pairs to learn text and video representations. Nevertheless, they do not pay enough attention to hard negative pairs and lack the ability to model different levels of semantic similarity. To address these two issues, this paper improves contrastive learning using two novel techniques. First, to exploit hard examples for robust discriminative power, we propose a novel Dual-Modal Attention-Enhanced Module (DMAE) to mine hard negative pairs from textual and visual clues. By further introducing a Negative-aware InfoNCE (NegNCE) loss, we are able to adaptively identify all these hard negatives and explicitly highlight their impacts in the training loss. Second, our work argues that triplet samples can better model fine-grained semantic similarity compared to pairwise samples. We thereby present a new Triplet Partial Margin Contrastive Learning (TPM-CL) module to construct partial order triplet samples by automatically generating fine-grained hard negatives for matched text-video pairs. The proposed TPM-CL designs an adaptive token masking strategy with cross-modal interaction to model subtle semantic differences. Extensive experiments demonstrate that the proposed approach outperforms existing methods on four widely-used text-video retrieval datasets, including MSR-VTT, MSVD, DiDeMo and ActivityNet.
Parameter-Efficient Tuning (PET) has emerged as a leading advancement in both Natural Language Processing and Computer Vision, enabling efficient accommodation of downstream tasks without costly fine-tuning. However, most existing PET approaches are limited to uni-modal tuning, even for vision-language models like CLIP. We investigate this limitation and demonstrate that simultaneous tuning of the two modalities in such models leads to multi-modal forgetting and catastrophic performance degradation, particularly when generalizing to new classes. To address this issue, we propose a novel PET approach called VioLET (Vision Language Efficient Tuning) that utilizes collaborative multi-modal gradients to unlock the full potential of both modalities. Specifically, we incorporate an additional visual encoder without learnable parameters and use these two visual encoders to compute the gradients of the context parameters separately. When conflicts arise, we replace the original gradient with an orthogonal gradient. Extensive experiments are conducted on few-shot recognition and unseen class generalization tasks using ResNet-50 or ViT/B-16 as the backbone. VioLET consistently outperforms several state-of-the-art methods on 11 datasets, showcasing its superiority over existing PET approaches. The code is available at https://github.com/Wang-Yaoming/VioLET.
In the Class-Incremental Learning (CIL) task, rehearsal-based approaches have received a lot of attention recently. However, storing old class samples is often infeasible in application scenarios where device memory is insufficient or data privacy is important. Therefore, it is necessary to rethink Non-Exemplar Class-Incremental Learning (NECIL). In this paper, we propose a novel NECIL method named POLO with an adaPtive Old cLass recOnstruction mechanism, in which a density-based prototype reinforcement method (DBR), a topology-correction prototype adaptation method (TPA), and an adaptive prototype augmentation method (APA) are designed to reconstruct pseudo features of old classes in new incremental sessions. Specifically, the DBR focuses on the low-density features to maintain the model's discriminative ability for old classes. Afterward, the TPA is designed to adapt old class prototypes to new feature spaces in the incremental learning process. Finally, the APA is developed to further adapt pseudo feature spaces of old classes to new feature spaces. Experimental evaluations on four benchmark datasets demonstrate the effectiveness of our proposed method over the state-of-the-art NECIL methods.
Zero-shot object detection (ZSD) aims to localize and recognize unseen objects in unconstrained images by leveraging semantic descriptions. Existing ZSD methods typically suffer from two drawbacks: 1) Due to the lack of data on unseen categories during the training phase, the model inevitably has a bias towards the seen categories, i.e., it prefers to subsume objects of unseen categories to seen categories; 2) It is usually very tricky for the feature extractor trained on data of seen categories to learn discriminative features that are good enough to help the model transfer the knowledge learned from data of seen categories to unseen categories. To tackle these problems, this paper proposes a novel zero-shot detection method based on a semantics-aware DETR and a class-wise adaptive contrastive loss. Concretely, to address the first problem, we develop a novel semantics-aware attention mechanism to mitigate the bias towards seen categories and integrate it into DETR, which results in a new end-to-end zero-shot object detection approach. Furthermore, to handle the second problem, a novel class-wise adaptive contrastive loss is proposed, which considers the relevance between each pair of categories according to their semantic description in order to learn separable features for better visual-semantic alignment. Extensive experiments and ablation studies on benchmark datasets demonstrate the effectiveness and superiority of the proposed method.