ACM MM 2024 论文列表

Existing pretrained text-to-video (T2V) models have demonstrated impressive abilities in generating realistic videos with basic motion or camera movement. However, these models exhibit significant limitations when generating intricate, human-centric motions. Current efforts primarily focus on fine-tuning models on a small set of videos containing a specific motion. They often fail to effectively decouple motion and the appearance in the limited reference videos, thereby weakening the modeling capability of motion patterns. To this end, we propose MoTrans, a customized motion transfer method enabling video generation of similar motion in new context. Specifically, we introduce a multimodal large language model (MLLM)-based recaptioner to expand the initial prompt to focus more on appearance and an appearance injection module to adapt appearance prior from video frames to the motion modeling process. These complementary multimodal representations from recaptioned prompt and video frames promote the modeling of appearance and facilitate the decoupling of appearance and motion. In addition, we devise a motion-specific embedding for further enhancing the modeling of the specific motion. Experimental results demonstrate that our method effectively learns specific motion pattern from singular or multiple reference videos, performing favorably against existing methods in customized video generation.

Spiking Neural Networks (SNNs) have received widespread attention due to their unique neuronal dynamics and low-power nature. Previous research empirically shows that SNNs with Poisson coding are more robust than Artificial Neural Networks (ANNs) on small-scale datasets. However, it is still unclear in theory how the adversarial robustness of SNNs is derived, and whether SNNs can still maintain its adversarial robustness advantage on large-scale dataset tasks. This work theoretically demonstrates that SNN's inherent adversarial robustness stems from its Poisson coding. We reveal the conceptual equivalence of Poisson coding and randomized smoothing in defense strategies, and analyze in depth the trade-off between accuracy and adversarial robustness in SNNs via the proposed Randomized Smoothing Coding (RSC) method. Experiments demonstrate that the proposed RSC-SNNs show remarkable adversarial robustness, surpassing ANNs and achieving state-of-the-art robustness results on large-scale dataset ImageNet. Our open-source implementation code is available at this https URL: https://github.com/KemingWu/RSC-SNN.

The advent of High Dynamic Range/Wide Color Gamut (HDR/WCG) display technology has made significant progress in providing exceptional richness and vibrancy for the human visual experience. However, the widespread adoption of HDR/WCG images is hindered by their substantial storage requirements, imposing significant bandwidth challenges during distribution. Besides, HDR/WCG images are often tone-mapped into Standard Dynamic Range (SDR) versions for compatibility, necessitating the usage of inverse Tone Mapping (iTM) techniques to reconstruct their original representation. In this work, we propose a meta-transfer learning framework for practical HDR/WCG media transmission by embedding image-wise metadata into their SDR counterparts for later iTM reconstruction. Specifically, we devise a meta-learning strategy to pre-train a lightweight multilayer perceptron (MLP) model that maps SDR pixels to HDR/WCG ones on an external dataset, resulting in a domain-wise iTM model. Subsequently, for the transfer learning process of each HDR/WCG image, we present a spatial-aware online mining mechanism to select challenging training pairs to adapt the meta-trained model to an image-wise iTM model. Finally, the adapted MLP, embedded as metadata, is transmitted alongside the SDR image, facilitating the reconstruction of the original image on HDR/WCG displays. We conduct extensive experiments and evaluate the proposed framework with diverse metrics. Compared with existing solutions, our framework shows superior performance in fidelity, minimal latency, and negligible overhead. The codes are available at https://github.com/pjliu3/MLP_iTM.

Traditional video steganography methods are based on modifying the covert space for embedding, whereas we propose an innovative approach that embeds secret message within semantic feature for steganography during the video editing process. Although existing traditional video steganography methods excel in balancing security and capacity, they lack adequate robustness against common distortions in online social networks (OSNs). In this paper, we propose an end-to-end robust generative video steganography network (RoGVSN), which achieves visual editing by modifying semantic feature of videos to embed secret message. We exemplify the face-swapping scenario as an illustration to demonstrate the visual editing effects. Specifically, we devise an adaptive scheme to seamlessly embed secret messages into the semantic features of videos through fusion blocks. Extensive experiments demonstrate the superiority of our method in terms of robustness, extraction accuracy, visual quality, and capacity.

Graph Neural Networks (GNNs) have proven effective in various scenarios. A key strategy involves pre-training existing graphs to extract knowledge that can be transferred to improve performance on downstream tasks, reducing the need for extensive labeled data. However, previous works commonly assumed that pre-training and fine-tuning occur in the same or closely related domains. A limitation is that for each individual graph without accessible pre-training data, a GNN must be trained from scratch, imposing high training overhead and hindering the ability of generalization. In this paper, we address the GNN multi-domain pre-training problem, which intends to pre-train a transferable GNN model from heterogeneous multi-source graph domains and then apply it in an unseen one with minor fine-tuning costs. To this end, we propose a scaLA ble Multi-source Pre-training (LAMP) method. For pre-training, LAMP presents a graph dual-distillation approach to distill massive knowledge from various graph domains to form synthetic homogeneous graphs. Simultaneously, high-level meta-knowledge from the synthetic graphs is extracted to train the GNN model, whose capability can be adjusted according to target graph contexts through a co-training modulation architecture. For fine-tuning, LAMP respectively aligns the target graph distribution, graph context, and graph task with the pretext so that the downstream task in the unseen domain can be reshaped to leverage the transferable knowledge efficiently. Extensive experiments on four different graph domain datasets show the superiority of LAMP.

We address image deraining under complex backgrounds, diverse rain scenarios, and varying illumination conditions, representing a highly practical and challenging problem. Our approach utilizes synthetic, real-world, and nighttime datasets, wherein rich backgrounds, multiple degradation types, and diverse illumination conditions coexist. The primary challenge in training models on these datasets arises from the discrepancies among them, potentially leading to conflicts or competition during the training period. To address this issue, we first align the distribution of synthetic, real-world and nighttime datasets. Then we propose a novel contrastive learning strategy to extract multi-view (multiple) representations that effectively capture image details, degradations, and illuminations, thereby facilitating training across all datasets. Regarding multiple representations as profitable prompts for deraining, we devise a prompting strategy to integrate them into the decoding process. This contributes to a potent deraining model, dubbed Rainmer. Additionally, a spatial-channel interaction module is introduced to fully exploit cues when extracting multi-view representations. Extensive experiments on synthetic, real-world, and nighttime datasets demonstrate that Rainmer outperforms current representative methods. Moreover, Rainmer achieves superior performance on the All-in-One image restoration dataset, underscoring its effectiveness. Furthermore, quantitative results reveal that Rainmer significantly improves object detection performance on both daytime and nighttime rainy datasets. These observations substantiate the potential of Rainmer for practical applications.

Transformer-based deep models for single image super-resolution (SISR) have greatly improved the performance of lightweight SISR tasks in recent years. However, they often suffer from heavy computational burden and slow inference due to the complex calculation of multi-head self-attention (MSA), seriously hindering their practical application and deployment. In this work, we present an efficient SR model to mitigate the dilemma between model efficiency and SR performance, which is dubbed Entropy Attention and Receptive Field Augmentation network (EARFA), and composed of a novel entropy attention (EA) and a shifting large kernel attention (SLKA). From the perspective of information theory, EA increases the entropy of intermediate features conditioned on a Gaussian distribution, providing more informative input for subsequent reasoning. On the other hand, SLKA extends the receptive field of SR models with the assistance of channel shifting, which also favors to boost the diversity of hierarchical features. Since the implementation of EA and SLKA does not involve complex computations (such as extensive matrix multiplications), the proposed method can achieve faster nonlinear inference than Transformer-based SR models while maintaining better SR performance. Extensive experiments show that the proposed model can significantly reduce the delay of model inference while achieving the SR performance comparable with other advanced models.

This paper explores sentence-level multilingual Visual Speech Recognition (VSR) that can recognize different languages with a single trained model. As the massive multilingual modeling of visual data requires huge computational costs, we propose a novel efficient training strategy, processing with visual speech units. Through analysis, we confirm that the visual speech units mainly contain viseme information while suppressing non-linguistic information. By using the visual speech units as the inputs of our system, we propose to pre-train a VSR model to predict corresponding text outputs on multilingual data. As both the inputs (i.e., visual speech units) and outputs (i.e., text) are discrete, we can greatly improve the training efficiency compared to the standard VSR training. Specifically, the input data size is reduced to 0.016% of the original video inputs. In addition, to stabilize the training, we apply curriculum learning where the inputs of the system begin with audio-visual speech units and gradually transition to visual speech units. After pre-training, the model is finetuned on continuous features. We set new state-of-the-art multilingual VSR performances by achieving comparable performances to the previous language-specific VSR models, with a single trained model.

Spiking neural networks (SNNs) have superb characteristics in sensory information recognition tasks due to their biological plausibility. However, the performance of some current spiking-based models is limited by their structures which means either fully connected or too-deep structures bring too much redundancy. This redundancy from both connection and neurons is one of the key factors hindering the practical application of SNNs. Although Some pruning methods were proposed to tackle this problem, they normally ignored the fact the neural topology in the human brain could be adjusted dynamically. Inspired by this, this paper proposed an evolutionary-based structure construction method for constructing more reasonable SNNs. By integrating the knowledge distillation and connection pruning method, the synaptic connections in SNNs can be optimized dynamically to reach an optimal state. As a result, the structure of SNNs could not only absorb knowledge from the teacher model but also search for deep but sparse network topology. Experimental results on CIFAR100, Tiny-imagenet and DVS-Gesture show that the proposed structure learning method can get pretty well performance while reducing the connection redundancy. The proposed method explores a novel dynamical way for structure learning from scratch in SNNs which could build a bridge to close the gap between deep learning and bio-inspired neural dynamics.

Large-scale point cloud semantic segmentation is a challenging task in 3D computer vision. A key challenge is how to resolve ambiguities arising from locally high inter-class similarity. In this study, we introduce a solution by modeling long-distance contextual information to understand the scene's overall layout. The context sensitivity of previous methods is typically constrained to small blocks(e.g. 2m x 2m) and cannot be directly extended to the entire scene. For this reason, we propose Long-Distance Context Modeling Network(LDCNet). Our key insight is that keypoints are enough for inferring the layout of a scene. Therefore, we represent the entire scene using keypoints along with local descriptors and model long-distance context on these keypoints. Finally, we propagate the long-distance context information from keypoints back to non-keypoints. This allows our method to model long-distance context effectively. We conducted experiments on six datasets, demonstrating that our approach can effectively mitigate ambiguities. Our method performs well on large, irregular objects and exhibits good generalization for typical scenarios.

We study an emerging and intriguing problem of multimodal temporal event forecasting with large language models. Compared to using text or graph modalities, the investigation of utilizing images for temporal event forecasting has not been fully explored, especially in the era of large language models (LLMs). To bridge this gap, we are particularly interested in two key questions of: 1) why images will help in temporal event forecasting, and 2) how to integrate images into the LLM-based forecasting framework. To answer these research questions, we propose to identify two essential functions that images play in the scenario of temporal event forecasting, i.e., highlighting and complementary. Then, we develop a novel framework, named MM-Forecast. It employs an Image Function Identification module to recognize these functions as verbal descriptions using multimodal large language models (MLLMs), and subsequently incorporates these function descriptions into LLM-based forecasting models. To evaluate our approach, we construct a new multimodal dataset, MidEast-TE-mm, by extending an existing event dataset MidEast-TE-mini with images. Empirical studies demonstrate that our MM-Forecast can correctly identify the image functions, and further more, incorporating these verbal function descriptions significantly improves the forecasting performance. The dataset, code, and prompts are available at https://github.com/LuminosityX/MM-Forecast.

Some recent methods address few-shot image classification by extracting semantic information from class names and devising mechanisms for aligning vision and semantics to integrate information from both modalities. However, class names provide only limited information, which is insufficient to capture the visual details in images. As a result, such vision-semantics alignment is inherently biased, leading to suboptimal integration outcomes. In this paper, we avoid such biased vision-semantics alignment by introducing CLIP, a natural bridge between vision and semantics, and enforcing unbiased vision-vision alignment as a proxy task. Specifically, we align features encoded from the few-shot encoder and CLIP's vision encoder on the same image. This alignment is accomplished through a linear projection layer, with a training objective formulated using optimal transport-based assignment prediction. Thanks to the inherent alignment between CLIP's vision and text encoders, the few-shot encoder is indirectly aligned to CLIP's text encoder, which serves as the foundation for better vision-semantics integration. In addition, to further improve vision-semantics integration at the testing stage, we mine potential fine-grained semantic attributes of class names from large language models. Correspondingly, an online optimization module is designed to adaptively integrate the semantic attributes and visual information extracted from images. Extensive results on four datasets demonstrate that our method outperforms state-of-the-art methods. The code is available at https://github.com/zhuolingli/FewVS.

Great progress has been made in rendering translucent materials in recent years, but automatically estimating parameters for heterogeneous materials such as jade and human skin remains a challenging task, often requiring specialized and expensive physical measurement devices. In this paper, we present a novel approach for estimating and transferring the parameters of heterogeneous translucent materials from a single 2D image to 3D models. Our method consists of four key steps: (1) An efficient viewpoint selection algorithm to minimize redundancy and ensure comprehensive coverage of the model. (2) Initializing a homogeneous translucent material to render initial images for translucent dataset. (3) Edit the rendered translucent images to update the translucent dataset. (4) Optimize the edited translucent results onto material parameters using inverse rendering techniques. Our approach offers a practical and accessible solution that overcomes the limitations of existing methods, which often rely on complex and costly specialized devices. We demonstrate the effectiveness and superiority of our proposed method through extensive experiments, showcasing its ability to transfer and edit high-quality heterogeneous translucent materials on 3D models, surpassing the results achieved by previous techniques in 3D scene editing.

As short-form video-sharing platforms become a significant channel for news consumption, fake news in short videos has emerged as a serious threat in the online information ecosystem, making developing detection methods for this new scenario an urgent need. Compared with that in text and image formats, fake news on short video platforms contains rich but heterogeneous information in various modalities, posing a challenge to effective feature utilization. Unlike existing works mostly focusing on analyzing what is presented, we introduce a novel perspective that considers how it might be created. Through the lens of the creative process behind news video production, our empirical analysis uncovers the unique characteristics of fake news videos in material selection and editing. Based on the obtained insights, we design FakingRecipe, a creative process-aware model for detecting fake news short videos. It captures the fake news preferences in material selection from sentimental and semantic aspects and considers the traits of material editing from spatial and temporal aspects. To improve evaluation comprehensiveness, we first construct FakeTT, an English dataset for this task, and conduct experiments on both FakeTT and the existing Chinese FakeSV dataset. The results show FakingRecipe's superiority in detecting fake news on short video platforms.

A soundscape is defined by the acoustic environment a person perceives at a location. In this work, we propose a framework for mapping soundscapes across the Earth. Since soundscapes involve sound distributions that span varying spatial scales, we represent locations with multi-scale satellite imagery and learn a joint representation among this imagery, audio, and text. To capture the inherent uncertainty in the soundscape of a location, we design the representation space to be probabilistic. We also fuse ubiquitous metadata (including geolocation, time, and data source) to enable learning of spatially and temporally dynamic representations of soundscapes. We demonstrate the utility of our framework by creating large-scale soundscape maps integrating both audio and text with temporal control. To facilitate future research on this task, we also introduce a large-scale dataset, GeoSound, containing over 300k geotagged audio samples paired with both low- and high-resolution satellite imagery. We demonstrate that our method outperforms the existing state-of-the-art on both GeoSound and the existing SoundingEarth dataset. Our dataset and code is available at https://github.com/mvrl/PSM.